proquest dissertations1+1 notice: library and archives canada published heritage branch 395...
TRANSCRIPT
1+1
NOTICE:
Library and Archives Canada
Published Heritage Branch
395 Wellington Street Ottawa ON K1A ON4 Canada
The author has granted a nonexclusive license allowing Library and Archives Canada to reproduce, publish, archive, preserve, conserve, communicate to the public by telecommunication or on the Internet, loan, distribute and sell theses worldwide, for commercial or noncommercial purposes, in microform, paper, electronic and/or any other formats.
The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission.
In compliance with the Canadian Privacy Act some supporting forms may have been removed from this thesis.
While these forms may be included in the document page count, their removal does not represent any loss of content from the thesis.
• •• Canada
Bibliotheque et Archives Canada
Direction du Patrimoine de !'edition
395, rue Wellington Ottawa ON K1A ON4 Canada
AVIS:
Your file Votre ref9rence ISBN: 978-0-494-57483-6 Our file Notre reference ISBN: 978-0-494-57483-6
L'auteur a accorde une licence non exclusive permettant a Ia Bibliotheque et Archives Canada de reproduire, publier, archiver, sauvegarder, conserver, transmettre au public par telecommunication ou par !'Internet, preter, distribuer et vendre des theses partout dans le monde, a des fins commerciales ou autres, sur support microforme, papier, electronique eUou autres formats.
L'auteur conserve Ia propriete du droit d'auteur et des droits moraux qui protege cette these. Ni Ia these ni des extraits substantiels de celle-ci ne doivent etre imprimes ou autrement reproduits sans son autorisation.
Conformement a Ia loi canadienne sur Ia protection de Ia vie privee, quelques formulaires secondaires ont ete enleves de cette these.
Bien que ces formulaires aient inclus dans Ia pagination, il n'y aura aucun contenu manquant.
Dominant patterns of interannual variability in the North Atlantic
St. John's
by
Farhana Nahed
A thesis submitted to the
School of Graduate Studies
in partial fulfillment of the
requirements for the degree of
Master of Science
Computational Science Program
Memorial University ofNewfoundland
August 2007
Newfoundland
Abstract
The low frequency variability of surface climate over the North Atlantic is described
using 55 years of observations from the Comprehensive Ocean-Atmospheric Data set and
NCEP reanalysis. Results are based on empirical orthogonal function analysis of sea
surface temperature (SST), air temperature, wind, heat flux, and precipitation. The
dominant spatial patterns of these parameters and sea level pressure are identified by
using canonical correlation analysis. It is shown that the zonal wind and air temperature
in the Labrador Sea are strongly correlated with the dominant modes of variability in the
atmospheric circulation. The SST anomalies in the Labrador Sea are weakly correlated to
the Sea Level Pressure (SLP) dominant modes of variability especially during the decadal
periods of warming in the 1950s and the 1990s. It is suggested that the long term
variability in the ocean circulation and sea ice at decadal time scale play a major role for
the SST in the Labrador Sea during these periods.
Acknowledgements
First I would like to thank my supervisor Dr. Entcho Demirov for his continuous
guidance and suggestions. He has been encouraging and always helpful. Without his
guidance and support this thesis would not have taken this final form.
I would also like to thank all the faculty and staff of Computational Science program and
Physics and Physical Oceanography Department for their assistance and support
throughout my program. I would also like to thank Chris Stevenson for being so helpful
with all of my computer problem.
Finally, I would like to thank Monowar and my friends, my Mom, Dad and others who
have contributed to my research and thesis.
11
Table of Contents
Abstract
Acknowledgements
Table of Contents
List of Figures
1 Introduction
1.1 Scales of variability of Climatic System: causes, scales and
i
ii
iii
v
implications..................... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 1
1.2 Implication of Climate Changes.......................................................... 5
1.3 The North Atlantic Oscillation ........................................................................ 6
1.4 Relations between atmospheric and the North Atlantic ocean
Variability................................................................................... 14
1.5 Problem description ........................................................................ 16
2 Methods of Data Analysis
2.1 Statistical Methods of Description and Understanding of Atmospheric and Ocean
data........................................................................................ 18
2.2 Empirical Orthogonal Functions ....................................................... 22
2.3 Canonical Correlation Analysis (CCA) ............................................... 25
lll
2.3.1 Canonical Correlation Patterns .................................................. 26
2.4 Computation of the Covariance Matrix ................................................ 30
2.4.1 Eigenvalues and Eigenvectors of a Square Matrix........................... 31
2.4.2 Data Structure..................................................................... 33
2.4.3 Algorithm ofEOF Analysis..................................................... 33
2.4.4 EOF Analysis Algorithm with Matlab ......................................... 35
2.5 Computation CC patterns after a transformation to EOF
Coordinates ............................................................................... 36
3 Seasonal, Interannual and lnterdecadal Variability of Atmospheric
Characteristics and Sea Surface Temperature in the North Atlantic
3.1 Seasonal Variability ....................................................................... 41
3.2 Inter-Annual and Inter-Decadal Variability ........................................... 50
4 Empirical Orthogonal Function Analysis of SST and Atmospheric Data
4.1 Dominant modes of variability of the North Atlantic atmospheric circulation
and NAO from 1948 to 2005 ............................................................ 60
4.2 EOF analysis of near surface atmospheric parameters and SST .................... 65
5 Canonical Correlation Analysis
5.1 Correlation Patterns of Variability in the North Atlantic ........................... 81
5.2 Correlation Patterns of Variability in the Labrador Sea ............................. 90
6 Conclusion and Future Works ...................................................... . 1 02
Abbreviations ................................................................................... . 104
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 106
IV
List of Figures
1.1 Idealized, schematic spectrum of Atmospheric Temperature ....................... .4
1.2 Satellite image of sea ice .................................................................. 5
1.3 North Atlantic oscillation index ........................................................... 8
1.4 Positive and negative NAO ............................................................... 1 0
3 .1 Seasonal mean air temperature .......................................................... 4 2
3.2 Seasonal mean zonal wind .............................................................. .44
3.3 Seasonal mean sea surface temperature .............................................. .47
3.4 Seasonal mean Total Heat Flux ........................................................ .49
3.5 Inter-annual and inter-decadal Variability of Air Temperature ..................... 51
3.6 Inter-annual and inter-decadal Variability of wind ................................... 54
3.7 Inter-annual and inter-decadal Variability ofSST ................................... 56
3.8 Inter-annual and inter-decadal Variability of heat flux .............................. 58
4.1 Dominant EOFs of SLP and their time series ofNA ................................. 64
4.2 Dominant EOFs of air temperature and time series ofNA .......................... 66
4.3 Dominant EOFs of air SST and time series ofNA ................................... 68
4.4 EOF analysis of SST (Deser & Blackmon) ........................................... 70
4.5 Time series of winter sea ice (Deser & Blackmon) .................................... 71
4.6 Dominant EOFs of wind and time series ofNA ........................................ 72
v
4.7 Dominant EOFs of precipitation and time series ofNA ............................... 74
4.8 Dominant EOFs of heat flux and time series ofNA ................................ 75
4.9 Difference of average heat flux winter month ofNA SLP EOF .................. 77
4.10 Difference of average thlST' I M ofNA SLP EOF (Cayan) .................... 78
5.1 Canonical correlation of SST in the North Atlantic ................................ 85
5.2 Canonical correlation of air temperature in the North Atlantic .................. 86
5.3 Canonical correlation of zonal wind in the North Atlantic ........................ 87
5.4 Canonical correlation of turbulent heat flux in the North Atlantic ............... 88
5.5 Canonical correlation of precipitation rate in the North Atlantic ................ 89
5.6 Canonical correlation of SST in the Labrador Sea ................................ 94
5.7 Water Transport Index .................................................................. 95
5.8 Canonical correlation of air temperature in the in the Labrador Sea ............. 98
5.9 Canonical correlation of zonal wind in the in the Labrador Sea ................. 99
5.10 Canonical correlation of turbulent heat flux in the Labrador Sea .............. 1 00
5.11 Canonical correlation of precipitation rate in the Labrador Sea ................ 1 01
Vl
1. INTRODUCTION
1.1 Scales of variability of Climate System: Causes, Scales and
implications
The climate has varied significantly and continuously on time scales ranging from years
to glacial periods and to the age of the earth. The variability of climate can be expressed
in terms of two basic modes: the forced variations, which are the response of the climatic
system to changes in the external forcing and the free variations due to internal
instabilities and feedbacks, leading to nonlinear interactions among the vanous
components of the climatic system.
The changes in the purely external factors that affect the climatic system, but are not
influenced by the climatic variables themselves, constitute what may be called the
external causes of climatic changes, whereas those changes that are related to nonlinear
interactions among the various physical processes in the internal system are called
internal causes. The distinction between the two classes of causes is not always very
clear.
The external causes comprise variations in both astronomical and terrestrial forcing. The
astronomical factors would include changes (a) in the intensity of solar irradiance; (b) in
the orbital parameters ofthe earth (eccentricity ofthe orbit, axial precession and obliquity
ofthe ecliptic); and (c) in the rate of rotation ofthe earth.
Among the terrestrial forcing we must consider (a) variations in atmospheric composition
(mixing ratios of carbon dioxide and ozone, aerosol loading, etc.) due to volcanic
eruptions and human activity; (b) variations of the land surface due to land use
(deforestation, desertification, etc.); (c) long-term changes of tectonic factors such as
continental drift, mountain-building processes, polar wandering, etc. Some other possible
terrestrial and astronomical forcing mechanisms are changes in solar output, the collision
of the earth with interplanetary matter, changes in volcanic activity, and changes in the
geothermal flux.
The internal causes are associated with many positive and negative feedback mechanisms
and other strong interactions between the atmosphere, oceans, and cryosphere. These
processes can lead to instabilities or oscillations of the system, which can either operate
independently or introduce strong modifications on the external forcing. Let us show by
some examples what we mean by the difference between an externally forced variation
and an internally free change. The seasonal or diurnal variations of the climate are clearly
related to the external astronomical forcing. But there are day-to-day weather variations
that take place independently of any changes in external forcing. These irregular
fluctuations with time scales of several days to a week may be connected with the
passage of migratory atmospheric perturbations (highs and lows on the weather map) or
with the passage of a frontal system. They are to be considered free, because they result
from the internal baroclinic instability of the zonal current which depends only on the
critical value of the latitudinal gradient of the temperature.
Figure 1.1 shows an idealized variance spectrum of the atmospheric temperature during
its past history as evaluated by Mitchell (1976). The analysis of the spectrum shows
several spikes and broader peaks. The spikes are astronomically dictated, strictly periodic
components of climate variation, such as the diurnal and annual variations and their
harmonics, whereas the broad peaks represent variations that are, according to Mitchell
(1976), either quasiperiodic or aperiodic however, with a preferred time scale of
2
energization. Many of these broad peaks cannot be directly explained by the known
external forcings. They indicate the existence of a strong free variability within the
system.
The peak at three to seven days is associated with the synoptic disturbances mainly at
mid-latitudes. The slightly raised region of the spectrum at 100-400 years is associated
with the "little ice age" that began near the early 17th Century with rapid expansion of the
mountain glaciers in Europe. The peak near 2500 years is perhaps due to the cooling
observed after the "climatic optimum," about 5000 years ago, which predominated during
the great ancient civilizations. The next three peaks are perhaps related to deterministic
astronomical variations of the orbital parameters of the earth, which are supposed to be
responsible for the ice ages (Milankovitch, 1941 ): ( a) the eccentricity of the orbit of the
earth with a cycle of about 1 00 000 years, (b) the axial precession with a cycle of around
22000 years, and (c) the change in the obliquity of the ecliptic or the axial tilt with a
period of about 41000 years. Finally, the peaks near 45 and 350 million years may,
according to Mitchell ( 197 6), be related to glaciations due to orogenic and tectonic
effects and to continental drift.
For a linear system the externally forced variations would lead to a simple relationship of
cause and effect: if the forcing is an oscillatory process the response of the system would
have exactly the same frequency. As we have seen, this is however, not always the case,
since the internal climatic system is inherently unstable and never reaches the equilibrium
state.
Climate variability results from complex interactions of forced and free variations
because the climate system is a dissipative, highly nonlinear system with many sources of
3
instabilities. The interactive and often nonlinear nature of the instabilities and the
feedback mechanisms of the climate system make it very difficult to obtain a
straightforward interpretation of cause and effect.
104 I I I I I I I I I I I T T
T SPECliUM SURFACE
103 t:-c -... " ;. .i = •• § ...
I 0 w
~ "D
u c z 10 2 E"
... §. c
-o( .
~
iii: ... o( " a ... c > ~ ~ §. ... w "i . ~I . c > ~
~ ... i= I ~ ~ :: j 10 5l
j I ... . ;. :: w a c~ D 0 ...
1111: E "D .. I £! ~Q
.., ... c I ... ;. ...
:::;:; ... 0
~ .. "D : :! ·:
~ 1t:- l
0.1 -1010 109 108 107 106 105 104 10 3 102 10 0.1 10-2 10-3 10-4
PERIOD IN YEARS
Figl.l: Idealized, schematic spectrum of atmospheric temperature between 1 o-4 and
1010yr adapted from Mitchell (1976)
4
1.2 Implication of Climate Changes
During the rece.nt two~three decades the climate change became a critical issue for the
annospheric and geophysical sciences. An important aspect of global changes is multi
and quasi-decadal, inter-annual climate variability. The North Atlantic is the only place in
the No11hcm Hemisphere where the atmosphere communicates deep oceanic water
masses through convective over-turning (falley 1984). Today we know that wanner
temperatures have been causing more and more ice to melt during summer in the
Nonl>ern Hemisphere. This change is linked directly to global wam1ing. In 2005 and
2006. the extent of winter icc was about 6% smaller than the average amount over the
past26 years. The retreat is also s ignificantly Larger than the Jong-tenn de<:reru;c of L .5%
to 2% in winter ice cover observed per decade over the same time period.
Fig 1.2: Tlu's S(IJel/ite image shows sea ice extent on II September 2006. The pink line
shows tlte ttverage ice extenf for Sep1en1ber, cn•eraged between 1979 and 2000 (Image:
NSIIX)
5
If sea surface temperature (SST) variations in the North Atlantic at the climatic timescale
are predicted it may be possible to improve the forecast of the global climate system.
Climate variability on decadal and longer time scales is a subject of increasing interest of
relevance. El Nino is perhaps the best-known example of the impact that changing sea
surface temperature has on our climate. Every three to seven years, this warming of
surface ocean waters in the eastern tropical Pacific brings winter droughts and deadly
forest fires in Central America, Indonesia, Australia, and southeastern Africa, and lashing
rainstorms in Ecuador and Peru. El Nino affects thousands of people worldwide, and
billions of dollars in economic impact. But changing SST patterns have broader
implications than just the El Nino and La Nina cycle.
In the Northern Hemisphere especially over the middle and high latitude during cold
month, the most prominent and recurrent pattern of atmospheric variability is the North
Atlantic oscillation (NAO). NAO refers to a redistribution of atmospheric mass between
the arctic and the subtropical Atlantic. NAO produces large changes in the mean wind
speed and direction over the Atlantic, the heat and moisture transport between the
Atlantic and the neighboring continents, and the intensity and number of storms, their
paths, and their weather. NAO has direct affect on agricultural harvests, water
management, and energy supply and demand.
1.3 The North Atlantic Oscillation
The North Atlantic Oscillation (NAO) is a climatic phenomenon in the North Atlantic
Ocean of fluctuations in the difference of sea-level pressure between the Icelandic Low
and the Azores high. Through east-west rocking motions of the Icelandic Low and the
6
Azores high, it controls the strength and direction of westerly winds and storm tracks
across the North Atlantic. It is related to and highly correlated with the so called Arctic
Oscillation.
The NAO is the dominant mode of winter climate variability in the North Atlantic region
ranging from central North America to Europe and much into Northern Asia. It is a large
scale seesaw in atmospheric mass between the subtropical high and the polar low. The
corresponding index (Fig 1.3) varies from year to year, but also exhibits a tendency to
remain in one phase for intervals lasting several years.
The NAO has a strong implication on the track of storms and depressions across the
North Atlantic Ocean and into Europe. The storm track exhibits variations from winter to
winter in its strength and position, but a particularly recurrent variation is for the storm
track to be either strong with a north-eastward orientation taking depressions into NW
Europe or weaker with an east-west orientation taking depressions into Mediterranean
Europe. Since the Atlantic storms that travel into Europe control the rainfall, there is a
strong influence on European precipitation patterns.
The year-to-year variability in storm tracks is associated with a change in the mean
atmospheric circulation averaged over the winter season. This variability is usually seen
in the anomalous sea level pressure (SLP) patterns associated with high or low NAO
winters. When the Iceland Low pressure centre is deeper than usual, the Azores High is
stronger than usual, and vice versa. The change in the mean atmospheric circulation
drives patterns of warming and cooling over much of the northern hemisphere. For
example, when the NAO index is high, the SLP gradient between Iceland and the
Azores/Iberia is enhanced, driving stronger westerly and southwesterly flow that carries
7
wann maritime air over the cold winter Eurasian land mass, bringing anomalously wann
winter temperatures.
Westerly wind<;: blo\o\~ng across the Atlantic, bring moist air into Europe. In years when
westerlies are strong, summers are cool. winters are mild and rain is frequent. If
westerlies are suppress..'(), the temperature is more extreme in summer and winter leading
to heat waves, deep freezes and reduced rainfalL
The permanent low·pressure system over fceland (the Icelandic Low) and the permanent
high-pressure system over the Azores (the Azores High) control the direction and
strength of westerly winds into Europe. A large dHlCrencc in the pressure at the two
stations (a high index year, denoted NAO+) leads to
N011h Atlantic Oscillation (NAO) index, 1864--2001 (Hurrell) J
l
0
-t
-1
ro m w w oo w ~ ~ ~ ~ ro w w ~ oo Fig l.J: North Atllmtic Oscillation Index
increased westerlies and. consequently, cool summers and mild and wet v.~nters in
Central Europe and its Atlantic fa~ade. In contras~ if the index is low (NAO-), westerlies
are suppressed, these areas sutTer cold v.~nters and s torms track southerly toward the
Mediterranean Sea. This brings increased stonn activity and minfall to southern Europe
and North Africa.
3
Especially during the months of November to April, the NAO is responsible for much of
the variability of weather in the North Atlantic region, affecting wind speed and wind
direction changes, changes in temperature and moisture distribution and the intensity,
number and track of storms.
Although having a less direct influence than for Western Europe, the NAO is also
believed to have an impact on the weather over much of eastern North America. During
the winter, when the index is high (NAO+ ), the Icelandic low draws a stronger
southwesterly circulation over the eastern half of the North American continent which
prevents Arctic air from plunging southward. In combination with the El Nino, this effect
can produce significantly warmer winters over much of the United States and southern
Canada.
The sea level pressure averaged over the winter is easier to measure than the storms
themselves, so the variability of the NAO can be measured by the difference between the
mean winter SLP at Gibraltar and the mean winter SLP over Iceland. Some people use
Lisbon or the Azores instead of Gibraltar, but it makes little difference.
The Positive NAO index phase shows a stronger than usual subtropical high pressure
center and a deeper than normal Icelandic low. The increased pressure difference results
in more and stronger winter storms crossing the Atlantic Ocean on a more northerly
track. This results in warm and wet winters in Europe and in cold and dry winters in
northern Canada and Greenland. The eastern US experiences mild and wet winter
conditions
9
Fig 1.4:/..efl Panel Positive Nor1h Atlantic Oscillation Righi Panel Negative North
Atlamic Oscillation.
The negative NAO index pha.o;e shows a weak subtropical lllg)l and a weak Icelandic low.
The rcduc(..-d pressure gradient results in fewer and weaker winter storms crossing on a
more wcst·east pathway. They bring moist air into the Mediterranean and cold air to
northen1 Europe. The US cast coast experiences more cold air outbreaks and hence
snowy weather conditions. Greenland, however, will have milder winter temperatures.
The NAO and its time dependence for instance appear central to the global change
debate. Surface temperatures over the Northern Hemisphere are likely wamler now than
at any other time over the past millennium (Mann et al .. 1999; Jones ct al., 2001) and the
rote of warming has been specially high over the past 40 years (Folland ct al .. 200 I;
Hansen et al.. 2002). A substantial fraction of this most recent \\tanning is linked to the
behavior of the NAO (Hurrell. 1996), in particular a trend in its index from large
tO
amplitude anomalies of one phase in the 1960s to large amplitude anomalies of the
opposite phase since the early 1980s. This change in the atmospheric circulation of the
North Atlantic accounts for several other remarkable alterations in weather and climate
over the extratropical Northern Hemisphere as well, and it has added considerably to the
debate over our ability to detect and distinguish between natural anthropogenic climate
changes. While it has long been recognized that the North Atlantic Ocean varies
appreciably with the overlying atmosphere (Bjerknes, 1964), another reason invigorated
interest in the NAO is that the richly complex and differential responses of the surface
layer, intermediate and deep layers of the ocean to NAO forcing are becoming better
documented and under stood (Visbeck et al, 1998). The intensity of winter time
convective renewal of intermediate and deep waters in the Labrador Sea and Greenland
Iceland-Norwegian Seas for instance, is not only characterized by large interannual
variability, but also by interdecadal variations that appear to be synchronized with
fluctuations in the NAO (Dickson et al., 1996). These changes in turn affect the strength
and character of the Atlantic thermohaline circulation and the horizontal flow of the
upper ocean, thereby altering the oceanic poleward heat transport and the distribution of
sea surface temperature.
On seasonal time scales, the upper North Atlantic Ocean varies primarily in response to
changes in the surface winds, air-sea heat exchanges and freshwater fluxes associated
with NAO variations (Cayan, 1992a,b). This does not mean, however, that the
extratropical interaction is only one-way. The dominant influence of the ocean on the
overlying atmosphere is to reduce the thermal damping of atmospheric variations, and
this influence becomes greater on longer time scales. The extent to which the influence of
11
the ocean extends beyond this local thermodynamic coupling to affect the evolution and
dynamical properties of the atmospheric flow is probably small, but the effect is non-zero
(Robinson, 2000; Kushnir et al., 2002).
New statistical analyses have revealed patterns in North Atlantic SSTs that precede
specific phases of the NAO by up to 9 months, a link that likely involves the remarkable
tendency of the extratropical ocean to preserve its thermal state throughout the year
(Kushnir et al., 2002.) On longer time scales, recent modeling evidence suggests that the
NAO responds to slow changes in global ocean temperatures, with changes in the
equatorial regions playing central role. (Hoerling et al., 2001). A second pathway that
offers hope for improved predictability of the NAO involves links through which changes
in stratospheric wind patterns might exert some downward control on surface climate
(Thompson et al, 2000). The NAO has a demonstrated influence on the physics,
hydrology, chemistry and biology of freshwater ecosystem across Northern Hemisphere
(NH) (Straile et al., 2003).
The spatial pattern related to NAO can be identified from the eigenvectors of the cross
covariance matrix,computed from the time variations of the grid point values of SLP or
some other climate variable. The eigenvectors, each constrained to be specially and
temporally orthogonal to the others, are then scaled according to the amount of total data
variance they explain. This linear approach assumes preferred atmospheric circulation
states come in pairs, in which anomalies of opposite polarity have the same spatial
structure. The largest amplitude anomalies in SLP usually occur during the boreal winter
months; however, throughout the year the leading pattern of variability is characterized
by a surface pressure dipole, and thus may be viewed as the NAO,although the spatial
12
pattern is not stationary (Barnston and Livezey, 1987; Hurrell and van Loon, 1997; Portis
et al., 2001 ). Since the eigenvectors are, by definition, structured to explain maximum
variance, it is expected that the "centers of action" of the leading EOFs will coincide with
the regions of strongest variability.
Most studies of the NAO focus on the NH winter months, when the atmosphere is most
active dynamically perturbations grow to their largest amplitudes. As a result, the
influence of the NAO on surface temperature and precipitation as well as on ecosystems
is also greatest at this time of the year.
In conclusion the North Atlantic Oscillation has a large climatic influence on the North
Atlantic Ocean and surrounding land masses. It is a major controlling factor in basic
meteorological variables such as surface wind, temperature and precipitation which have
large socio-economic impacts on energy, agriculture, industry, traffic and human health
throughout the whole of Europe and eastern North America.
Early in this century, scientists aware of the NAO also became interested in its socio
economic impacts (Meinardus, 1905). When more recently the NAO came into the focus
of scientific work again, a number of studies considered the influence of this large scale
variability pattern on local meteorological quantities such as seasonal mean precipitation,
temperature and wind velocity (e.g. Wallace and Gutzler, 1981; Deser and Blackmon,
1993; Hurrell, 1995; Rogers, 1990; Malberg and Bokens, 1997; Ulbrich and Christof,
1999).
13
1.4 Relations between atmospheric and the North Atlantic Ocean
variability
There has been an intensification of efforts toward describing and understanding climate
variability in the North Atlantic Ocean, in the past decade. A significant fraction of that
variability is associated with the NAO. While the NAO state may be characterized by an
index of that pressure difference (e.g., Hurrell 1995), the NAO is so strong mode that
alternative indices based on surface air temperature differences (van Loon and Rogers
1978), storms (Rogers 1997), sea ice (Deser et al. 2000), sea surface temperature (SST)
patterns (Deser and Blackmon 1993; Kushnir 1994; Sutton and Allen 1997), or SST
gradients (Czaja and Marshall 2001), when regressed on SLP, yield much the same
dipole SLP anomaly patterns.
When averaged for the winter months, the SLP-based index (Hurrell 1995) exhibits
interannual to decadal and interdecadal fluctuations. High and low states of the NAO
index reflect enhanced and diminished midlatitude westerlies, with related shifts in wind
patterns and intensities of air-sea exchanges of heat (Cayan 1992a,b ), freshwater (Hurrell
1996), and momentum (Visbeck et al. 1998). Since the general oceanic circulation,
including upper ocean gyres and thermohaline overturning, is wind and buoyancy forced,
there is little surprise in finding oceanic patterns in the observational record that are
coordinated with the low frequency components of atmospheric variability. Notable are
SST anomalies that develop directly as a time-integrated response to persistent NAO air
sea heat flux anomalies (Bjerknes 1964; van Loon and Rogers 1978; Deser and
Blackmon 1993; Kushnir 1994). If the NAO SLP-SST -heat content fluctuations were the
only signals of coordinated atmosphere--{)cean variability, then that association would be
14
amenable to a null hypothesis explanation: that the ocean integrates atmospheric forcing
to redden the spectral response relative to the forcing spectrum (Hasselmann 1976;
Frankignoul and Hasselmann 1977). Battisti et al. ( 1995), for example, hindcast winter
SST for the North Atlantic using observed atmospheric circulation anomalies and an
ocean simulated as a field of one-dimensional mixed layers on the top a diffusive deep
ocean (no advection). The relative success of this and similar SST hindcasts for large
areas of the North Atlantic has given considerable weight to the idea of mixed layer
integration of surface heat flux as the leading mode of extratropical ocean-atmosphere
interactions.
The upper ocean's mean advective-diffusive heat balance involves a large annual surface
heat flux from the midlatitude ocean into the overlying atmosphere, supported by oceanic
heat transport convergence into the region. Thus SST anomalies are more than merely
one-dimensional mixed layer responses to local atmospheric forcing; they reflect
departures of the heat balance from the climatological annual mean and seasonal cycles.
These departures can lead to interannual persistence of winter SST anomalies through a
reemergence mechanism (Alexander and Deser 1995) whereby a winter mixed layer
anomaly is seasonally sequestered beneath a summer thermocline and then reappears in
the subsequent winter.
The observations indicate that this mechanism acts both locally (viewing as an Eulerian
field) and in a Lagrangian framework. Underlying the decadal evolution of the observed
oceanic heat content patterns are not only the thermal inertia of winter mixed layer
dynamics but also modulations of the ocean gyre circulation.
15
1.4 Problem Description
The objective of this thesis is to evaluate the impact of the variability of the atmospheric
circulation on the Labrador Sea. This evaluation is done through analysis of available
data from observations and atmospheric reanalysis. One major goal of this work is to
assess if the variability of atmospheric near surface parameters and SST of the Labrador
Sea is predictable by using information about the SLP only. The later provides
information about the major modes of atmospheric circulation and their change in time.
The thesis includes analysis of the relation between SLP modes and variability of the
Labrador Sea based on the traditional methods of Empirical Orthogonal Function (EOF)
and Canonical Correlation Analysis (Chapter2).
The interannual variability of the North Atlantic is discussed in Chapter 3. Chapter 4
presents results from EOF analysis of the studied prameters. The discussion in Chapter 4
defines the major characteristics and events of the North Atlantic variability during the
recent 50 years.
Chapter 5 discusses the connection between the variability of the atmospheric circulation
and the interannual change of the near surface atmospheric parameters and SST in the
Labrador Sea.
16
Chapter 2
2. Methods of Data Analysis
This chapter describes the methods used in the analysis of long term variability in the
North Atlantic SST and atmospheric characteristics. The approach is based on the use of
statistical methods for analysis of simultaneous multivariate data. The aim of the study is
to describe the dominant patterns of spatial variability about the long term mean. This can
be accomplished through use of Empirical Orthogonal Functions (EOF) and Canonical
Correlation Patterns (CCA) analysis.
2.1 Statistical Methods of Description and Understanding of
Atmospheric and Ocean Data
The climate is a complex dynamical system and its variability is influenced by a great
number of external factors. In climate studies only small portion of these factors can be
considered, while the rest are usually considered as background noise. Correspondingly
we consider every climate characteristics X as a sum of two components
X=Xd +X'
Here X d is a deterministic part of X which is the subject of our study. X' is non
deterministic, stochastic component of X' which is not correlated to X d •
17
The statistical approach has certain drawbacks due to the fact that small variations in
ocean and atmospheric characteristics which usually are considered as noise some times
can have significant impact on the large-scale variability. Ocean vertical mixing, for
instance, which has spatial scales of centimeters or millimeters, may have significant
impact on the large scale circulation. This feedback between the large and small scales is
due to the nonlinear dynamics of the climate system. Nonlinear components of the
hydrodynamic equations include important advective terms. The thermodynamic part also
contains various nonlinear processes. The nonlinearities make the climate system
unpredictable beyond certain characteristic times. These characteristic time scales are
different for the climate subsystems such as the ocean, atmosphere, lithosphere and
cryosphere. The climate subsystems also can interact in a complex way influencing each
other on different time scales. The interactions often have the character of positive and
negative feedbacks. One important feedback, which is a main interest of this study is the
feedback between SST and atmospheric circulation. The surface momentum and
buoyancy fluxes drive the ocean circulation. The ocean currents transport heat and salt
polewards. This transport heats the poles and makes the ocean surface in the polar and
subpolar areas warmer. At the same time the ocean surface temperature influences back
the atmospheric circulation. The feedback between the atmosphere and ocean processes is
observed at different scales in variability of the characteristics of the two sub-systems at
different scales. One example is ElNino and Southern Oscillation (ENSO) which is
observed in the results simultaneous variability in ocean and atmospheric characteristics
in the tropical Pacific Ocean at interannual time scale.
18
Another example is the North Atlantic Oscillation (NAO) which dominates atmospheric
circulation variability over the North Atlantic at a decadal time scale. While there is
strong evidence that NAO dominates the variability of atmosphere over the North
Atlantic and therefore the atmospheric forcing of ocean circulations, the importance of
SST to the atmosphere is not widely accepted
In this study, we assume that the long term sea surface temperature and atmospheric
characteristics over the North Atlantic Ocean consist of two components, one which is
related to the interaction between atmosphere and ocean over the North Atlantic and
second, which is stochastic. The latter is due to error in observations or calculation of the
ocean and atmospheric parameters, or is related to processes not observable in the
available data. EOF and CCA techniques are used in this study to identify patterns of
simultaneous variability in the atmosphere and surface layer of the North Atlantic Ocean.
Let X 1 represents certain atmospheric or oceanic characteristics at some vertical level.
X 1 is a vector of dimension N=m*n which contains values of the 2-dimensional
characteristics with dimension n in latitude and m in longitude. The mean state
expectation of X 1 is ji = E(X). Statistical parameters have considerable complexity in
the climatological context. In particular, the computed mean is not entirely reliable as an
estimate of the climate system's true long term mean state. The computed mean usually
contain errors caused by taking observations over a limited observing period, at a discrete
times and a finite number of locations. It may also be affected by the presence of
instrumental, recording, and transmission errors. The climatological mean should be
understood also to be a moving target because today's climate is different from that
which prevailed thousands years ago.
19
The time mean of the characteristics X 1
is denoted by p . Then the variability of X 1 IS
described in terms of anomalies X' = X 1
- p . The anomalies at two different points
often have a tendency to vary jointly. This 'co-variability' may be quantitavely described
by the covariance matrix Lxx = E(x ;x ;T )where E is the expectation operator. In
statistics covariance indicates the strength and direction of a linear statistical relationship
between two random variables. When the linear relationship is stronger, then the
covariance also becomes higher. The time mean (u) and covariance matrix I only
describe statistical properties of important class of random processes. A random variable,
or a random process, is said to be stationary if all the statistical parameters are
independent of time. As a result, parameters such as the mean and variance, also do not
change over time or position. Most statistical techniques assume that the observed
process is stationary.
Most climate parameters that are sampled more frequently that one per year are not
stationary but cyclo-stationary, simply because of the seasonal forcing of the climate
system. Long-term averages of monthly mean sea-level pressure exhibit a marked annual
cycle. The mean annual cycle is computed as monthly mean(fi(m),m = 1, .... 12). The
anomalies remaining after subtracting the monthly mean are:
In the following, the studied atmospheric and ocean characteristics will be assumed that
they are cyclo stationary. Two methods are used in this study to identify the major modes
of interannual variability. First EOF analysis is used to identify the dominant pattern in
20
anomaly fields of atmospheric and ocean characteristics. The canonical correlation
analysis is then applied in a study of the interrelation between basic modes of variability.
2.2 Empirical Orthogonal Functions
Lets consider a stochastic characteristic X with mean fi . Assume that the anomalies X;
K
can be expanded in to a finite series x; = Lai,t~i with time dependent coefficient iii,/
and fixed patterns ~~ (Empirical Orthogonal Functions) and K = m * n . The variance of
the time coefficients iii,t usually decreases quickly with increasing index i, so that good
K'
approximations X; ~ L iii it are usually possible for K' << n * m. The patterns are i=l
chosen to be orthogonal so that optimal coefficients iii,t are obtained by simply projecting
the anomalies X; onto the patterns~~ . The coefficients iii are the EOFs coefficients.
The first step of the EOF analysis is to find one pattern e1, such that
E I = E cllx - (X 'e I )e 1112
) is minimized. Here the scalar product IS defined as
where * denotes complex conjugate, and the norm isllall =~(a, a). This first step finds
the projection of the random vector X onto one-dimensional subspace spanned by the
fixed vector e 1 • The variance of X which is contained in this subspace is:
21
El = E(llxll2
- 2(x ,e1) * x 1e1 + (x ,e1 ) * (x ,e1
))
=E(IIxll2 -(x,el)*(x,el))
= var(X)- var((x ,e1)) (1)
where the variance of the random vector X is defined to be the sum of variances of the
elements ofX and T denotes matrix or vector transpose. It can be shown that
Var((x ,e1) )= e11 Lxx e1 where I is the covariance matrix of X. The solution for the
first EOF vector e1, which minimizes £ 1 =. (IIX -(X,e1)e1
11
2
) under the constraint
[[e 1[[ = 1 can be found by using method of the Lagrange multipliers with the constrain
[[e 1[[ = 1. The minimization of equation (1) leads to:
(2)
where t denotes complex conjugate of transposed vector. This equation means that the
first EOF is the eigenvector e1 of the covariance matrix Lxx eigenvalue A..
To find the second EOF we have repeated the same procedure by minimizing
The result is that e2 is the eigen vector of Lxx that corresponds to its second largest
eigenvalue /L 2 • This second pattern is orthogonal to the first because the eigenvectors of
a Hermitian matrix I xx are orthogonal to one another.
22
We can continue in this way to define the following EOFs. The result of this process may
be formulated in the following way:
Let X be an m-dimentional random vector with mean jl and covariance matrix Lxx. Let
A1 ~ A2 ~ ••• ~Am be the eigen values of Lxx and let e1 , ••• ,em be the corresponding
eigenvectors of unit length. Since Lxx is the Hermitian, the eigenvalues are non-negative
and the eigenvectors are orthogonal.
(i) The k eigenvectors that correspond to A,, ... , Ak minimize
k
(ii) Ek= Var(X)- LA; i=l
m
(iii) Var(X) =LA; i=l
The EOF coefficients or principal components a;= (x ,e;) =X 1 e;• = e1r X are
uncorrelated and hence independent. The covariance of a; and a 1
satisfies the relation:
cov(ai'a1 ) = E( ((x- fi }ei )((x- fi }e' )*) -~TE{I(X- -xx- -)' \_. 1 -iT"- j 1 -iT- f Q =e ~ -f.l, -p, r =e L..,.e =Aje e =
Usually major part of the variance of X can be represented by the first few EOFs. If the
original variable has m components the approximation of X by a= (ap ... ,ak) with
k' << K. Using only limited number of EOFs leads to a significant reduction of amount
23
of analyzed data while retaining most of the variance. It is possible to clearly associate
the first EOF with a known physical process but this is much more difficult to do with the
second EOF because it is constrained to be orthogonal to the first EOF. Additional
analysis is needed in order to connect the computed EOFs with the underlying dynamics.
This analysis is usually based on the existing knowledge about the process and/or
applications of statistical methods like CCA.
The basic hypothesis of the EOF analysis is that studied characteristics X may be split
into two parts - deterministic X d and stochastic noise X ' . In the EO F analysis we need to
distinguish between the EOFs which correspond to Xd and toX'. The former are the
ones, which are going to be used in pattern analysis.
After computing the EOF analysis, one needs to select the number of dominant EOFs,
which spawn the space of X d • The most popular procedure to separate physically relevant
and noise related EOFs is based on so called Rule N The latter is based on the
assumption that the phase space (spanned by all EOFs) can be splitted into two subspaces
one which contains only noise, and second, which contains dynamic variability. It is
assumed that the dynamical sub-space is spanned by the EOFs with large and well
separated eigen values. Another approach is suggested by North et al., (1982) which is
based on estimation of typical errors for i; and eigenvectors ~;
The North's "Rule-of-Thumb' states that
24
If the sampling error of a particular eigenvalue ~A is comparable to or larger than the
spacing between A, and a neighboring eigenvalue, then the sampling error ~e of the
EOFwill be comparable to the size ofthe neighboring EOF.
2.3 Canonical Correlation Analysis (CCA)
Canonical Correlation Analysis is used to study the joint variability of random vectors
- - - I - I X andY. The objective of CCA is to find a pair of patterns I x and I r so that the
correlation between linear combinations X T] x 1 and Y r] r
1 is maximized. A second pair
-z -z -r-z - -z of patterns I x and I r is found so that X I x and Y T I r are the most strongly
- - - T- I correlated linear combinations of X andY that are not correlated with X I x and
-r- 1 Y I r , and so on.
25
2.3.1 Canonical Correlation Patterns
Let us consider a m x dimensional random vector X and a my dimensional random
vector Y. It is required to find a m x dimensional vector j x and a my dimensional
vector j Y such that the inner products flx = (X, j x) and flY = ( Y, j Y) have maximum
correlation. i.e.:
p = Cov(flx ,flY)
~Var(flx )Var(flY) - T - - -lx Cov(X,Y)Iy
(1)
where flx and flr are called canonical variates. If a pair of vectors lx and IY
- -maximizes the above equation then all vectors ax I x and a Y I Y do the same for any
nonzero ax and ay . Thus I x and I Y can be arbitrary normalized. We can choose the
patterns such that
where I xx and I YY are the covariance matrices of X andY. The above equation can
be rewritten as p- f-T " 7 - X ~XY}y ( 4) where I xr is the cross covariance
matrix Ixx = E((x- it X x-v- jly )). Similarly to the approach used in previous section
26
vectors fx and are found by max1m1zmg
(5)
where s and 1J are Lagrange multipliers that are used to account for constraints (2) and
(3). Setting the partial derivatives of E to zero, we get
(6)
(7)
and
8 E T - -- = '"" f + 21]'"" + = 0 a]y L....xx X L..,.yylY (8)
which is equivalent to
(9)
- -Then (9) is substituted into (7) to obtain a pair of eigen-equations for fx and fr
(10)
(11)
The eigenvectors of the two matrices are related to each other through a simple equation:
if fx is a solution of equation (1 0) then I~~ I :r fx is a solution of equation (11)
provided that their joint eigenvalue is nonzero. Finally equation (4) is maximized by
- -letting fx and fr be the solution of equations (1 0) and (11) that corresponds to the
27
largest eigenvalue A= 4r;1J. We have got the canonical random variables fJx =\X, 1x)
and flY = \ Y, 1Y) that are most strongly correlated, the natural next step is to find the
value ofp. Using equation (4), (6), (8) and (2), (3) in sequence we find:
P2
= 1~ Lxr1Y1~"L:r1x = 41Jr; 1/Lxx1x1~Lrr1Y =A
In this way we have found that the correlation is the square root of the eigenvalue that
- -corresponding to eigenvectors fx and /y .
To obtain m =min( m x, my) pairs of patterns ( 1;, 1;) and m corresponding pairs of
canonical variates the derivation detailed above can be repeated.
(12)
(13)
with correlation P; = Cov(fJ;x ,fJ;Y) =A, i-1,2, ... ,m.
The canonical variates and patterns are indexed in order of deceasing eigenvalue A;. Pairs
of canonical variates are uncorrelated (i * j).
Cov ( fJ/, fJ/ ) = Cov(/]1Y, PI) = Cov ( /3/, PI ) = 0
- -Assume X and Y are of the same dimension of m. Then the canonical variates
jjx =(f31x, ... ,f3:,)' and jJY =(fJt, ... ,fJ~)'can be viewed as the result of coordinate
transforms that have been applied to X and Y . The transformations relate jjx and jJY to
28
To find F X we need jjX = ((x, n ), ... ,(x,f;) y = f~ X where fx is the m X m matrix
-. with eigenvector f~ in its ith column. Thus
Cov(x,jjx )= Cav(X,f~X)= Cav(x, x)fx = Lxx fx
Substituting equation (14) for X, we get
The columns of F x and F v .F ~ and F ~ , are called the canonical correlation patterns.
The canonical variates /3;x and flr are called also canonical correlation coordinates
are normalized to unit variance, the canonical correlation patterns are expressed in the
units of the field they represent and they indicate the typical strength of the mode
represent by the patterns.
29
2.4 Computation of the Covariance Matrix
The correlation matrix [R] is introduced as a device for compactly representing the
mutual correlations among K variables. The computation begins with the ( n x K) matrix
[X] of data values whose correlations are to be computed. Each row of this matrix is a
vector, consisting of one observation each of K variables. The number of these rows is
the same as the sample size, n, since each of these row vectors is a sample point (in K
dimensions). Thus, [X] is essentially just an ordinary data table. An individual data
element xi,k is the ith observation of the kth variable.
Define the (n * n) matrix [1 ], whose elements are all equal to 1. The (n x K) matrix of
anomalies (in the geophysical sense of variables with their means subtracted), or centered
data [X'] is then
[X']= [X] _ _!_[l][X] n
(1)
The second term in Equation (1) is a (nx K)matrix containing the sample means. Its n
rows are all the same, and each consists of the K sample means in the same order as the
corresponding variables appear in each row of [X]. Multiplying [X'] by the transpose of
itself, and dividing by n - 1, yields an estimation S of the covariance matrix L xx
[Lxxl=-1 [x'Y[x']
n-1
The covariance matrix [Lxx] is a symmetric (K x K) matrix whose diagonal elements
are the sample variances of the K variables, and whose other elements are the co variances
30
among the K variables. For example, s1,1 is the sample variance of the first variable, and
s1,2 is the sample covariance between the first and the second variables. Because
variances are so often expressed in terms of covariance matrices, it is fairly common to
see the notation S1
,1
for the sample variance, which is equivalent to the more common and
familiar s,2 •
2.4.1 Eigenvalues and Eigenvectors of a Square Matrix
The formal definition of an eigenvalue of a matrix A is that it is any value A, which is a
root of the characteristic equation of the matrix,
det( A - A * J) = 0
A is an eigenvalue of A if and only if there is a nonzero vector x, known as an
eigenvector (or sometimes a "right" eigenvector), with the property that
A*x=A*x
Note that there must also be a "left" eigenvector y, with the property
y * A = A'* y = A* y
The characteristic equation has exactly N roots, so a matrix has N eigenvalues. An
important consideration is whether any eigenvalue is a repeated root, which determines
how hard the eigenvector computation will be.
If a matrix has the maximum possible number of linearly independent eigenvectors
(namely N, the order of the matrix), then the eigenvalues and eigenvectors can be used to
diagonalize the matrix. This only happens when the matrix is normal.
31
A nonzero vector x is an eigenvector of the square matrix A if
A*x=lt*x
for some scalar value It, called the associated eigenvalue.
Sometimes this eigenvector is more particularly described as a right eigenvector, so that
we may also consider left eigenvectors, that is, vectors y for which it is true that
y * A = A '* y = J.1 * y for some scalar J.1 •
For every eigenvalue of a matrix, there is at least one eigenvector. Every nonzero
multiple of this eigenvector is also an eigenvector. If, and only if, an eigenvalue is a
repeated root, then there may be more than one linearly independent eigenvector
associated with that eigenvalue. In particular, if an eigenvalue is repeated 3 times, then
there will be 1, 2 or 3 linearly independent eigenvectors corresponding to that eigenvalue.
For many statistical applications, eigenvalues and eigenvectors are calculated for real (not
containing complex or imaginary numbers), symmetric matrices. Eigenvalues and
eigenvectors of such matrices have a number of important and remarkable properties,
some of which also hold for the eigenvalues and eigenvectors of square matrices more
generally. The first of these properties is that the eigenvectors of symmetric matrices are
orthogonal. That is, their dot products with each other are zero. Furthermore, since each
eigenvector is defined to have unit length, the dot product of any eigenvector with itself is
one:
T {l,i = j X X =
I J O,i ::j:. j
32
2.4.2 Data Structure
Let us assume that we have measurements of some variable at locations
(ApqJ1),(A2 ,qJ2 ) ••• (Ap,<f'p) taken at times fpt 2 , •• .tn where AK is latitude and <f'K is
longitude of kth points. For each times t 1 (j = 1, ... n) we can think of the p -dimensional
vector F1; (i = l, ... p) as one map of the field .We store this measurements in a matrix F as
n maps each being p points long. So one row of matrix F is one map for time t 1
• We
arrange each map into row vector in F with size p. We can then interpret each of the p
columns in F as a time series for a given location. The EOF analysis is performed using F
as the data matrix.
2.4.3 Algorithm of EOF Analysis
Let us assume that we have removed the mean from each of the p time series in F, so
that each column has zero mean. We form the covariance matrix ofF by calculating
R=FTF, and then we solve the eigenvalue problem. RC = CA
A is a diagonal matrix containing the eigenvalues A; of R . The c; column vector of C
are the eigenvectors of R corresponding to eigenvalues A; . Both A and C are the size
p x p. For each eigenvalue A; chosen we find the corresponding eigenvector C;. Each of
this eigenvectors can be regarded as a map. These eigenvectors are the EOFs we are
looking for. In what follows we always assume that the eigenvectors are ordered
according to the size of the eigenvalues. Thus, EOFI is the eigenvector associated with
33
the biggest eigenvalue, and the one associated with the second biggest eigenvalue is
EOF2, etc. Each eigenvalue A; , gives a measure of the fraction of the total variance in R
explained by the mode. The value in percentage is found by dividing the A; by the sum
of all the other eigenvalues (the trace of A).
The eigenvector matrix C has the property that CTC=CCT=I or the EOFs are
uncorrelated over space. Another way of stating this is to say that the eigenvectors are
orthogonal to each other.
The pattern obtained when an EOF is plotted as a map, represents a standing oscillation.
The time evolution of an EOF shows how this pattern oscillates in time. To see how the
first EOF evolves in time we calculate:
The n components of the vector a1 are the projections of the maps in F on EOFl, and
the vector is a time series for the evolution ofEOFl. In general, for each calculated
EOFj, we can find a corresponding a1
. These are the expansion coefficients ofthe EOFs.
Just as the EOFs where uncorrelated in space, the expansion coefficients are uncorrelated
in time.
A common use of EOF is to reconstruct a cleaner version of the data by truncating this
sum at some j = N << p; that is, we only use the EOFs of the first (largest) few
eigenvalues or the most significant EOFs. The rational is that the first N eigenvectors
are capturing the dynamical behavior of the system and the other eigenvectors
(corresponding to the smallest eigenvalues) are just due to random noise. This
assumption does not always have to be true.
The algorithm for finding EOF is:
34
• First form matrix . from the observations, and remove the time mean of each
time series.
•
•
•
Then find the covariance matrix R=FTF .
Then find the eigenvalues and eigenvectors of R by solving RC = CA
Afterwards find the biggest eigenvalues and their corresponding eigenvectors, the
EOFs.
• Then find the expansion coefficients by calculating aj=FCj
2.4.4 EOF analysis Algorithm with Matlab
Matlab algorithm was developed for EOF analysis that includes the following steps.
• Define a matrix M, with each row as one map, and each column a time series for a
given station. That means in matrix M, the data need to be put in such a way that
row represents map and column represents time series.
• Remove the mean of the column.
• Form the covariance matrix by the command: R=F'*F;
• Calculate the eigenvalues and eigenvectors of the covariance matrix.
• Find the expansion coefficients corresponding to eigenvalue number i.
• Compute the vector giving the amount of variance explained for each eigenvalue.
35
2.5 Computation of CC patterns after a Transformation to EOF
Coordinates
In the calculation of CCA, if the data are transformed into EOF space before the analysis
the CCA algebra becomes considerably simpler. Suppose that only the first k x and
kr EOFs are retained, so that
(20)
We have used the renormalized versiOns of the EOFs and their coefficients
d -iT ( ~ )I I 2 - i an e = /l,; e . The CCA IS then applied to the random
- T - T vector X' = ( a1x +, •.. , a{,+) and Y' = (a{+, ... , a[y+) . An advantage of this approach is that
it is often possible to use only the first few EOFs.
Another advantage is that the algebra of the problem is simplified since Lxx· and
Ln" are both identity matrices. Thus according to the equations (10, 11) f~.and f~. are
T T eigenvectors of Lxr· Lxr· and Lxr· Lxr· respectively. As these are non-negative
definite symmetric matrices, the eigenvectors are orthogonal. Moreover, the canonical
correlation patterns F ~· = i ~· and F ~· = i ~· .
A minor disadvantage is that the patterns are given in the coordinates of the renormalized
EOF space. To express the pattern in the original coordinate space it is necessary to
reverse transformation (20) with (18) and (19):
36
The canonical correlation patterns are no longer orthogonal after the above back
transformation and f i- F.
37
Chapter 3
Seasonal, Interannual and Interdecadal Variability of Atmospheric Characteristics and Sea Surface Temperature in
the North Atlantic
The physical processes in the ocean surface layer play an important role in formation of
the atmosphere and ocean variability. The incident solar radiation absorbed by the ocean
heats the surface layer. The ocean surface is in contact with the atmosphere and
exchanges heat and water with the near surface air masses.
Seasonal and latitudinal variations in ocean sea surface temperature are driven primarily
by variations of insolation. The amount of solar radiation incident on the ocean surface
depends on the latitude, season and time of the day. The mean solar flux per unit area
perpendicular to the solar beam at the mean position ofEarth is equal to S0 = 1367 w/m2.
The Earth is approximately spherical; and the planets surface is heated differentially. The
solar heat flux at the ocean surface depends on the local solar zenith angle ()8 at each
latitude. It is equal to zero at the equator and approaches 90° at high latitudes. Beyond the
astronomical factors, the solar radiation incident to the ocean surface depends also on
atmospheric characteristics. The solar heat enters the atmosphere is partly reflected back
to the space and partly absorbed by the air molecules. The amount and spectrum of the
solar radiation at the ocean surface depends therefore on the content of greenhouse gasses
and clouds in the atmosphere.
38
The solar heat flux per unit area in the upper atmosphere is
where d is the mean distance from which the flux density S0 is measured, and d is the
actual distance from the sun. The solar radiation Q0 incident on the ocean surface is large
in the equatorial areas and smaller in the polar ocean. The ocean reflects back part of the
solar heat Q0 incident to its surface. The amount of the reflected radiation depends on the
roughness of the ocean surface. The strongest reflection is observed in the ocean areas
covered by ice. The part of the radiation reflected by the ocean surface is called albedo
as , which for the major part of the ocean has a value a 8 ~ 0.3. In the polar ocean as
may have values higher than 0.3.
The ocean emits back to the atmosphere outgoing long wave radiation R,, . It is greatest
over the tropical ocean, where cloud cover is relatively low. R,, is lowest in the polar
regions and in regions of persistent high cloudiness in the equatorial areas.
The zonal averaged annual mean net radiation
Rnet = Qo(l-a,)- RL
is positive equator-ward of 40° of latitude and negative pole-ward of that latitude.
Assuming that the ocean and atmosphere are in steady state conditions, which means that
long term annual mean ocean and atmospheric characteristics do not change, one can
write the balance of the energy at the ocean surface as
Rnet -LE-SH -Mea =0
39
where LE is the latent heat, SH the sensible heat, /!:,.Feo is the horizontal flux out of the
column of ocean below the ocean surface. We will often refer to !!:.Feo also as the
meridional heat flux. Mea is due to the horizontal circulation and turbulent mixing in
the ocean.
The meridional heat transport exists in both atmosphere and ocean. If this transport would
not exist, the tropics would be much warmer and the poles much colder. The net turbulent
heat flux.
HT =LE+SH
is the heat exchange through the ocean surface due to processes of turbulent mixing in the
boundary layers of atmosphere and upper ocean.
The near surface atmospheric and ocean characteristic depend in a complex way on the
surface energy balance. At the same time some of them like SST, atmospheric
temperature, humidity, wind speed influence the intensity of the heat exchange through
the surface. In this way, the near surface atmospheric and ocean characteristics may be
considered as important indicators for the long-term changes in state of the atmosphere
and ocean. This section presents analysis of interannual variability in the near surface
atmospheric characteristics and SST.
Following the previous studies ofKushnir (1994), Deser and Blackmon (1993) the winter
SST and atmospheric parameters are studied here. During the winter the convective
processes intensify the vertical turbulent mixing in the surface ocean layer. The depth to
which the vertical transport of heat reaches during this season varies usually from about
lOOm in mid-latitudes up to about 2000m in the deep convection areas. During the rest of
the year the vertical mixing is limited to a relatively thin surface layer.
40
In this chapter first the seasonal mean characteristics of SST, atmospheric temperature,
zonal wind and net turbulent heat flux are described. The seasonal variability usually is
much stronger than variability at interannual and interdecadal time scales. The later
however is important because it shows the tendencies in the ocean and atmospheric
variability at long time scales. The winter anomalies of SST and near surface atmospheric
characteristics are discussed in section 2.
3.1 Seasonal Variability
This section describes the seasonal variability of atmospheric characteristics and SST.
The seasonal mean parameters are calculated for winter December, January, February
(DJF), spring March, April, May (MAM), summer June, July, August (JJA) and autumn
September, October, November (SON) seasons. As described in the previous section the
near surface atmospheric characteristics and SST distribution depends on the radiative
fluxes, interaction between the atmosphere an ocean and atmospheric dynamics (see Fig
3.1 ). The dependence of the absorbed solar radiation on the latitude results in quasi-zonal
structure of atmospheric parameters i.e. their variability in zonal direction is smaller than
in meridional. The air temperature equatorward of 30lN, where the net radiative balance
is positive, is relatively high and remains between 20° and 30° during the whole year. In
the subpolar areas the seasonal mean air temperatures change from -5 to+ 10 degrees C.
41
a) Wonlef b) Spri!lll
c) Summer d) Aulllmn
0 • 10 " •
Fig 3.1 :Seasonal Mean Air Temperature (0C).(a)Winter,(b)Spring,(c)Summer.(d)Autumn
42
There is also a longitudinal change in the atmospheric temperatures during all seasons. In
all seasons the maximum of air temperature is in the western part of the tropical ocean.
This is the area of Gulf Stream, which transports warm waters along the North America
Coast. The Gulf Stream heat transport is a major contributor to the meridional ocean heat
transport in the subtropical North Atlantic. The atmospheric temperature in eastern
subtropical North Atlantic is colder than in western and central part of the basin. This is
an area of intensive upwelling, where cold waters are transported from the deep ocean to
the surface. The relatively cold ocean waters in the upwelling areas near the eastern coast
of the Sub-tropical North Atlantic cool the near surfaces air.
The atmospheric temperature differs also in the eastern and western part of the sub-polar
North Atlantic. As in the sub-tropics, this difference is due to the transport in the ocean.
The ocean current along the western coasts of the sub-polar North Atlantic transport cold
waters from the polar areas towards the Labrador Sea. This currents transport also sea-ice
which melts in spring in the Labrador Sea. The melting of the ice causes additional
cooling and freshening of the surface waters. The temperature in the eastern sub-polar
North Atlantic is influenced by the Gulf Stream, which transports warm waters, towards
the polar areas. The seasonal atmospheric temperatures here are 1 0° to 15° higher than in
the western part of the sub-polar North Atlantic.
The atmospheric zonal wind (Fig 3.2) over the North Atlantic is influenced by two major
elements of the atmospheric circulation. These are the passats in the tropical area
43
Fl'g 3.2: Seasonal Mean Zonal Wind (m/s). a) Winter b) Spring c) Summer d) Autumn
••
and westerlies in the mid-latitudes. The passat winds in the tropical area are
predominantly easterlies. They persist with some seasonal variability during the whole
year. The later is driven by the upward atmospheric flow in equatorial area. Here humid
air masses warmed by the ocean surface move upward. At high altitudes the dominant
flow is poleward, where the uplifted at the equator air masses, propagate towards the
tropics. The passats form the lower branch of Hadley circulation, and they bring dry air
masses towards the equator. Passats have a strong westward component, which is present
at low latitudes during all seasons (Fig 3.2 a, b, c, d). In summer the Hadley cell
intensifies in the Northern Hemisphere. Correspondingly the passats have their strongest
seasonal mean zonal component during this season (Fig 3.2c).
The westerlies in the mid and sub-polar latitudes persist during all seasons. This is the
area of polar jet stream. It forms between latitudes 30~ and 70~. The strong eastward
flow at high altitudes is formed due to the meridional gradient of atmospheric
temperature and is a consequence of the thermal wind relation, which governs the large
scale geophysical flows. Additionally there is a tendency of developing cyclonic
disturbances in mid-latitudes along the jet stream to form fronts. The frontogenesis
processes related to the dynamics of cyclones, result in concentration of north-south
temperature gradients into relatively narrow regions. This contributes to the sharpness of
the jets. The area of positive zonal wind in mid-latitudes (Fig 3.2) is the zone where
intensive cyclones and related frontal systems pass the North Atlantic. During the winter,
the cyclones bring relatively cold and dry air masses over the North Atlantic.
Additionally the cyclonic atmospheric eddies are related with strong near surface winds.
As discussed in the beginning of this chapter, under these conditions the surface heat and
45
water fluxes intensify the areas where the contrast between atmospheric and ocean
temperature is high. The process of cyclogenesis is stronger in winter and autumn.
Correspondingly these are the seasons of strongest westerlies and related storms passing
the mid latitudes of North Atlantic. Strong easterlies are present along the eastern coast of
Greenland. They have strongest magnitude in winter and autumn.
The SST, as the atmospheric temperature has a well defined zonal distribution. The
equatorial and tropical areas of the North Atlantic are warm during the whole year. The
temperature here varies between 20° to 25°C with maximum in summer when the
absorbed solar radiation is higher. The minimum of the SST during all seasons is in the
sub-polar areas. Deviation from the zonal structure of SST distribution is observed in the
areas of western boundary currents- Gulf Stream, the Labrador Current and East
Greenland Current.
The Gulf Stream and its extension, the North Atlantic current, transports warm waters
towards the Northeastern part of the North Atlantic. The SST along the North American
coast and the western coast of Europe is high due to this transport. The Labrador Current
and the eastern Greenland current transport cold waters toward the coast of North East
North Atlantic. This flow, which during the winter and spring transports also sea-ice, has
the strongest impact on the SST during these seasons. Another area of deviation from the
zonal distribution of SST is in the North Western Africa upwelling region.
46
o 2 4 s a 10 12 14 18 18 ~ n
Fig 3.3: Seasonal 1\lean SST(°C). a) JY;nter. b) Spr;ng. c) Summer. d) Alllunm
47
The spatial and temporal variability of TA and SST have similar patterns. One reason for
this is the surface turbulent heat transport. The direction of the surface turbulent heat
transport across the surface is always towards the cooler component of the system ocean
atmosphere, which keeps TA and SST close.
Net heat flux Qnet consists of two radiative flux components, which are the surface
shortwave and longwave radiation and two components of the turbulent heat: sensible
and latent heat fluxes. Qnet over the oceans is one of the key parameters governing the
atmosphere-ocean interaction. It is required for weather prediction and climate studies
related to the global energy balance and water cycle. Using suitable models, net heat flux
can be used to reproduce and predict variations of the global ocean regime and its impact
on atmospheric dynamics, as well as variations in regional hydrological processes and
water resources and responses to such environmental changes as the increase in
greenhouse gas forcing. Here we discuss the seasonal variability of turbulent net heat flux
Hr . Its dominant component is the evaporative heat loss. The sensible heat exchange at
the ocean surface is smaller, except over the warm Gulf Stream.
H 7 is predominantly negative in the whole North Atlantic. In the tropics there is an area
of strong heat loss by the ocean, which coincides, with the zonal belt of passats winds.
This is the area of strong evaporation and negative latent heat flux. The turbulent heat
flux in the Gulf Stream area is negative during all the seasons. The heat loss by the ocean
here is strongest in winter when the contrast between the atmospheric and sea
temperatures is relatively large. Hr has high seasonal variability also in the Labrador and
Greenland seas.
48
Willet Sprilg
.... .,.. .. ., 0 "
Summer Autumn
Fig 3.4: Seosonol Meon Toto/ Heot Flux (JY!m1) (a) Winter, (b)Spring, (c) Summer, (d)
Armmm
49
These are two areas of deep water formation, which are strongly cooled during the winter
(Fig 3.4 a,b ). In summer the heat loss through the turbulent exchange here is relatively
weak.
3.2 Inter-Annual and Inter-Decadal Variability
In this section winter anomalies of atmospheric characteristics and SST are discussed.
The anomalies are calculated for winter months only and then averaged for every year.
The annual mean anomalies are then used to describe the interannual variability of the
North Atlantic.
As mentioned in the previous section the SST and near surface atmospheric temperature
are quasi-zonaly distributed. This fact was used in previous studies (see Kushnir,1994) of
the North Atlantic interannual variability, where the anomalies were studied separately in
several zonal belts of the basin. In the previous section it was shown that there are two
major areas in the North Atlantic which have different characteristics and dynamics: 1)
the subtropical North Atlantic is characterized by high SST, TA and wind (passats) which
have westward direction 2) the mid and high latitudes with colder atmosphere, ocean
surface and winds with predominant eastward direction. Here we study SST and
atmospheric characteristics averaged for these two regions.
50
a) PO!i>l11160 -11160
c) Period 1996 - 2005 .. OA
0
~· ...
b) Period 1970- 1960
.. 0
~· _,
-u
""'f,...L,--,ce,..c--::1001',--,ce,,.::--::1910=--,ce,.::---:-.,..~-,,.i
Fig 3.5: Inter-Annual and lntcr-Dccadal Variability of Air Temperature
51
Fig 3.5.d shows the air temperature anomaly averaged within two zonal belts,
20° ~ rp ~ 45° and 45° ~ rp ~ 70°. Here after, the zonal belt 20° ~ rp ~ 45° will be shortly
referenced as the southern area, and the zonal belt 45° ~ rp ~ 70° as the northern area.
The surface atmospheric temperature anomaly distributions in the two areas (Fig 3.5 d)
show some similarities. The decadal mean anomalies are predominantly positive in the
both regions in early 1950s and negative in the 1970s. The temperature anomalies in the
two areas (southern and northern) show a warming trend at the end of the study period
from 1995 to 2005. In the southern region (blue line) this trend starts in the late 1980s
while in the northern region (green line) it begins later- in the second half of the 1990s.
The spatial distribution of the atmospheric temperature anomaly averaged for three
periods of time : (a) from 1950 to 1960, (b) from 1970 to 1980 and (c) from 1995 to 2005
are shown on Fig 3.5 a, b, c. The first period 1950-1960 (Fig 3.5 a) was relatively warm
in the Western North Atlantic. The pattern of the highest atmospheric temperature
extends along the Gulf Stream area, the Labrador Sea and off east coast of Greenland.
Negative atmospheric temperature anomalies are observed in the zonal belt between 20~
and 30~ and along the Western North Africa and Western European coasts.
The cold atmospheric temperatures during the 1970s dominate in a zonal region of the
North Atlantic between 30~ and 55~, in the Northern Labrador Sea and along the
Northwestern Africa coast. Slightly positive anomalies are observed in the rest of the
North Atlantic. The comparison of Fig 3.5 with Fig 3.2 suggests that the mid-latitude
areas of strongest warming in the 1950s and cooling in the 1970s approximately
coincides with the mid-latitude area of westerlies. In the tropical areas where the
52
dominant winds are passats, the anomalies were negative in the 1950s and close to zero in
the 1970s. In the late 1990s the atmospheric anomalies are positive almost in the whole
North Atlantic with the exception of a small region in the Caribbean basin.
The wind anomalies in the northern and southern areas are clearly anticorrelated during
the period from 1948 to 1995 (Fig 3.6 d). From 1950-1965 the zonal westerlies in the
northern area tend to decrease, while the zonal wind anomaly increases in the southern
sub-region. During the rest of the period until 2005, the wind interannual variability is
opposite in the two regions. The reason for this anticorrelation between winds anomalies
in the sub-regions, which is discussed in more details in the next chapter, are related to
the variability of the North Atlantic Oscillation (NAO).
The zonal air mass transport is an important factor for the air-sea interaction over the
North Atlantic. Intense zonal winds bring relatively dry air from the continent over the
ocean. The weakening of the zonal winds off the coast ofNewfoundland in the 1950s is
related to weakening of storms and cold air masses transport in this area.
Correspondingly the winter of 1950s were warmer than usual in this part of the North
Atlantic. In opposite the westerlies were stronger than usual in 1970s (Fig.3.6.b). The
winters of 1970s were correspondingly colder than usual at mid-latitudes (Fig.3.5.b).
However, the comparison of (3.5c) and (3.6c) shows that this relation between the zonal
wind and TA anomalies is not always present.
In the 1990s the zonal wind anomalies in the whole North Atlantic show overall increase
in the intensity of both system- westerlies in the mid and polar latitude, and easterlies in
53
a)Period 1960 ·1960 b) Period 1911) ·1960
d) Zonal Wind anomaly c) Period 1995 • 2005 3.---~--~--------------~----,
2 2
t t
0 • ·1 ·1
Fig 3.6: lntcr·Annualand lntcr·Decadal Variability of Wind
54
the tropics. One would expect that the intensified transport of continental cold and dry air
would intensify the heat loss in the both areas. TA (Fig 3.5) and SST anomalies (see
below) however are both positive over the main part of the basin. This presumably may
be explained by the fact that there are other factors beyond the zonal transport, which
influence the surface atmospheric temperature distribution during this period. Two of the
most important factors which remain beyond this study are radiative fluxes and
meridional transport. They are the major 'candidates' which may be responsible for the
overall warming of the North Atlantic.
Fig 3.7 of SST anomalies clearly show presence of three periods in the North Atlantic
SST variability. The SST anomalies are in general positive in the 1950s and negative in
the 1970s. After 1995, the SST anomaly increase until the end of the period which is
2005. The patterns of SST in the 1950s, 1970s and in the 1990s are very similar with the
pattern of TA , suggesting that the variability of these two parameters is strongly
connected.
The net heat flux (Fig 3 .8) exhibits more complex variability than the parameters
discussed so far. The net heat flux depends in a complex way on TA, SST and wind
speed. The variability of all these parameters together with atmospheric humidity
influences the net heat flux. In the 1950s the net heat flux is predominantly positive in the
almost whole North Atlantic except the area of Gulf Stream and North Atlantic current.
55
a) Period 1950 -19«10 b) Period 1970-1980
01 .(1.8 .0.4 .0.2 0
d) Sea Sulface T~re ....,aly c) Period 1995-2005
0 0
OJ .0.& .(1,4 .0.2 0 0.4 0.8 0.8
f'ig 3.7: Inter-Annual and lnter-De.:adal Variability of SST
56
In the 1970s the heat flux is predominantly negative with exception of the Gulf Stream
area and a area off the coast of Western Europe. In the 1990s the heat flux anomaly in
tropical and subtropical North Atlantic is positive. In the areas of deep-water formation in
the Labrador and Greenland Sea the heat flux is negative. In the northern Labrador Sea,
there is an area of positive heat flux anomaly, which however is relatively small. The
spatial mean of the heat flux anomaly in the 2000 (Fig 3.8 a) is positive in the southern
and negative in northern area.
The analysis of seasonal mean and anomalies of the surface atmospheric characteristics
and SST suggest that there are distinct patterns of ocean-atmosphere relationship
associated with interannual variability. The middle and high latitude SST and TA display
a long term fluctuations with positive anomalies in the 1950s and 1990s and negative
anomalies in the 1970s. The zonal wind anomalies are negatively correlated in middle
and low latitudes. The patterns in the zonal wind show some relation with the SST and
TA anomalies in the mid latitudes in the 1950s and in the 1970s. In the 1990s this relation
is less evident. In this period the turbulent net heat flux is predominantly negative in the
polar areas and positive in the low and middle latitudes.
57
a) Period 1960 • 1960 b) Poriocl1970-1980
• 0 10 " 20
c) Period 1995 • 2005
0 0
~ ·15 .to 0 10 20 .,,.. ..., "" "" "'' ... .,., .,f
Fig 3.8: Inter-Annual and lnter-Decadal Variability of Total Heat Flux
58
Chapter 4
Empirical Orthogonal Function Analysis of SST and Atmospheric Data
The aim of Empirical Orthogonal Function (EOF) analysis is to determine the principal
modes of variability of the studied data. Empirical Orthogonal Function analysis is a
decomposition of a signal or dataset on to orthogonal basic is of functions, which are
determined from the data. EOF method finds both time series and spatial patterns (see
Chapter 2). Once the eigenvectors are found, it is important to separate those
eigenvectors (or EOFs) which contain data that represent physically relevant patterns
from those that represent only noisy variations. It has also been noticed that in analysis of
this type, the eigenvalues fall exponentially and most of the variability is captured only
by very few vectors.
The data used in this study are time series of monthly mean anomalies of SST and
atmospheric data in the North Atlantic and Labrador Sea for winter months (DJF) of the
period between 1948 and 2005. The SST fields are taken from COADS data set
(Woodruff et al., 1987) and have a horizontal resolution of 2 degrees. The atmospheric
parameters are from National Center for Environmental Predictions (NCEP). The
analysis covers the North Atlantic sector north of 20~. At each grid point the anomalies
have been calculated by subtracting the long term monthly mean.
59
4.1 Dominant modes of variability of the North Atlantic atmospheric
circulation and NAO from 1948 to 2005
The first EOF of sea level pressure (SLP) of North Atlantic, explains 44% of the total
variance present in the original data. The large positive values centered around latitude
45° N and the large negative values centered around 65° N are indicative of two regions
whose mean sea level pressures are generally inversely related. This EOF defines a
pattern which is a well-known low-frequency atmospheric circulation pattern called the
North Atlantic Oscillation (NAO). The NAO is characterized by large-scale SLP
variability associated with a subtropical high I polar low system over the Northern
Atlantic. During a positive NAO, (positive values on Fig 4.1 b) the subtropical high is
stronger than usual and the polar low is deeper than usual. The increased pressure
gradient causes stronger winter storms to cross over the Atlantic. During a negative
NAO, the subtropical high and polar low are both weaker than usual, resulting in fewer I
less severe storms crossing the Atlantic.
The intensity of the NAO is usually studied by using the so-called NAO index, which is
the normalized difference of pressure anomalies over Iceland and Portugal. This
characteristic can be calculated for the last 200 years, because SSP data for these two
stations are available for this period of time. The SSP analysis data, which we use in
these calculations, are available for not more than 50 years. Though such data do not
allow analyzing a long enough sequence of temporal variability of SSP, they provide
much better information about the spatial variability of SSP than the two point
observations (Portugal, Iceland) only.
60
There is a wide range of atmospheric and ocean variability attributed to the changing
NAO, which have the potential to cause change in the marine environment. In particular
in many studies during the recent two decades, it was shown that processes of water mass
formation in the North Atlantic are related to the NAO. The radical interannual and
interdecadal changes in the production of the convectively formed mode waters of the
West Atlantic, which Dickson et al.(1996) suggest to be a part of a coordinated pan
Atlantic pattern of convective activity, driven by the changing NAO. From the late 1950s
to 1970, a regime of cold winter air temperatures and extreme snowcover greatly
enhanced the land-sea temperature gradient at the US eastern seaboard, spinning up more
storms than normal offshore (Dickson and Namias, 1976; Hayden, 1981). The local effect
was to focus the center of maximum storm activity off the US eastern seaboard where the
cold stormy conditions caused maximum cooling the ocean surface (see Fig 3.5b and
3.7b) and formation and ventilation of the Eighteen-Degree. The remote effect was to
reduce storminess over the Labrador Sea, so that Labrador Sea convection became
increasingly suppressed and fresh water built up at the surface (Lazier, 1980, 1988,
1995). Labrador Sea Water (LSW) production resumed abruptly in winter 1971-1972
with a rapid removal of the surface fresh water accumulation by the vertical spreading as
the cold winter regime ended at the US east coast and intense, chill northwesterlies and
storminess returned to the Labrador Sea. Later the tendency has been towards
intensifying and deepening ventilation of the Labrador Sea, with a progressive cooling
and freshening of LSW during the period of the 1990s. The LSW density increased as
convection began to produce cold but saline sublayer of North Atlantic Deep Water
(Dickson et al., 1996). This variability in the LSW formation is related to the changes in
61
the atmospheric characteristics driven by the NAO. From the 1950s the EOF1 coefficient
(Fig 4.1b) has decreased from positive to its lowest (negative) value for the whole period
1948-2005, which occurred in the 1960s. This period was the period of low storm activity
in the North Atlantic and was characterized by high Atmospheric and surface ocean
temperatures (Fig 3.5, Fig 3.7).
The connection between the ocean variability and NAO was however different in the
1990s. It is known that the NAO had a tendency to be positive since the late 1980s. This
is the period when NAO had its highest intensity (see Fig 4.1b) for the whole period
1948-2005. The SST during this period (see Fig 3.7a) showed a strong tendency towards
increasing. This tendency started earlier (in 1990) in the southern region (southern of 45°)
and later (in 1995) in the northern region. The low correlation ofNAO and SST suggests
that beyond the direct air-sea interaction, there was probably another factor, which
influenced the SST variability in the 1990s. As mentioned in introduction of Chapter 3 an
important factor for the thermal regime of the North Atlantic is the ocean circulation. The
long term variability of the ocean heat transport however remains a challenging problem
due to the lack of data. The most important contributors to the meridional transport are
Gulf Stream and North Atlantic Current.
As Curry and McCartney (200 1) point out, the main North Atlantic Current is driven by
the gradient of potential energy anomaly (PE) across the mutual boundary between the
subtropical and subpolar gyre. Since P E reflects the vertical density structure and heat
content of the upper ocean to well below the wind-driven layer, it follows that
coordinated changes of opposite sign in the production and characteristics of the mode
waters in each gyre will have the potential to drive deep-seated changes in the P E
62
gradient, and hence in the strength of Atlantic gyre circulation. If these changes in the
density and heat content of mode waters are attributable to the NAO, then the
amplification of the NAO to extreme values over the past three or four decades is likely
to have been followed by a corresponding multidecadal spin-up of the Atlantic gyre
circulation.
From the observed PE differences between the centers of the two gyres, Curry and
McCartney calculated that the long term increase in the NAO index to the mid 1 090s was
accompanied by a 30% increase in the east going baroclinic mass transport along the
gyre-gyre boundary. Both subpolar and subtropical gyres contributed equally to the
changes in their transport index. Thus in response to the NAO, the North Atlantic gyre
circulation during the early 1990s is likely to have been at its strongest for more than a
century.
The second EOF of SLP over North Atlantic exhibits 27% variance, which is called the
Eastern Atlantic (EA) pattern (Barnston and Livezey, 1987). The EA is related to an
atmospheric pattern with low-pressure center at about rp = 55° N and A = 25° W, which
is inversely related to the SLP in the rest of the North Atlantic.
63
(b) Ttne 111111es SI.P EOF 1 (a)Sea Surface Pressure EOF 1 variance: 0.44487
1960 1970 2000
(d) Time series SLP EOF 2 (c)Sea Surface Pressure EOF 2 variance: 0.27048
0.5
0
~.5
·1 ~.08 ~.07 ~.06 ~.05 ~.04 ~.03 ~.02 ~.01 0 0.01
1990 2000 2010
Fig 4.1: Dominant EOFs ofSea Level Pressure and their time series (red curve is the
filtered time series) of North Atlantic.
64
4.2 EOF analysis of near surface atmospheric parameters and SST
EOF Decomposition is applied to Air Temperature of North Atlantic (Fig 4.2). The first
structure explains 51% of the variance, the second structure 13%, and the third EOF
explains 6%. The first EOF has triple structure with positive values of the US coast and
opposite negative sign in the Labrador Sea and tropical North Atlantic. The time
variability (Fig 4.2b) follows closely the NAO index.
We have compared our results with the results of Deser and Blackmon (1993 ). The first
two EOFs of air temperature are similar with the pattern of SST EOF, just the order of
mode is reversed. The second EOF of air temperature, which accounts for 13% of the
variance, exhibits maximum amplitude off the Newfoundland, North Labrador Sea and
off the coast of east Greenland. The time series of EOF2 is dominated by along warming
trend from the 1950s to the 1970s, followed by a cooling trend during 1970s and 2000s.
65
(c)--.--~ 2 (ooo0.1S167)
•M•r------~"'!-~ _., ... _, ·--:·---~---
•
-~·r~------!,.~-~"-~-~---.o;---co"~
"
~l .,_ ' ·-
• ·•
-
. ·-
-
. ··~
. ·- . ·- -=·~--=
til,_- .. • .. , - eooo•
: '
- - - ·- -Fig 4.2: Dominant EOFs of air temperature ant/their time series (red curve is the
filtered time series) of North Atlantic.
66
-
The dominant SST is shown on Fig 4.3. The first normalized eigenvalue is 0.23, the
second eigenvalue is 0.18, and the third eigenvalue is 0.14. Recall that normalized
eigenvalues represent the fraction of variance explained by the structure associated with
that eigenvalue. Therefore, the first EOF structure ofNorth Atlantic Ocean explains 22%
ofthe variance, the second structure 17%, and the third EOF explains 13%.
The SST dominant EOFs define patterns, which are pretty similar to the EOFI and EOF2
of the heat flux (see below). The EOFI identifies two areas of opposite variability. The
first is southeastern of Newfoundland and Labrador Sea and second is the subtropical
North Atlantic and Gulf of Mexico. This structure is very close to the structure of the
EOF2 of heat flux. The time variability of EOFI is different from the EOFs discussed so
far. Though it shows some variability at interannual time scale, there is a stronger
changes in this EOF at decadal time scale. The coefficient of this EOF is predominantly
positive from 1950 to 1970, negative from 1970-1990 and positive in 1990s.
The second EOF defines a triple with two areas in subtropical and subpolar North
Atlantic, where SST anomalies due to EOF2 are opposite to those in Gulf Stream Area.
This pattern is known (see Cayan, 1992) to be strongly correlated with the NAO.
Correspondingly the time variability of the EOF2 coffecient vary in a similar way as the
EOFI ofSLP.
67
00 s--..r.c.T.....,....,.eOF 1 (cr •02214) .. • • • -~
-~,,_,_-,,._,..-,,c._..---...- . ·-. ·- . - -· .. .,..._ _ _. ___ IIO<i't
.. • • • ...
.. Q'IT-._., ____ _
... •
• • ... .. ,_ ·- ·- - ·- ·- - -
Fig 4.3: Dominant EOFs o/SSf and their lime series (red cun•e is lhe filtered lime
series) of North Allanlic
Deser and Blackmon (1993) have done an EOF analysis of SST in the North Atlantic.
They have found the first EOF of SST, which accounts for 45% of the variance, has
uniform polarity over the entire basin. The largest loading occurs along the Gulf Stream.
The time series of EOF1 SST exhibits a sudden transition from below normal values to
above normal values around Second World War. This EOF was not found in the present
study and also in the work of Cayan (1992).The technique for measuring SST was not
accurate before early 1940s. As also discussed by Deser and Blackmon, the EOF1 is
probably related to the change of technique of SST observation.
The second EOF which accounts for 12% of the variance, exhibits a center of action east
of Newfoundland at the boundary between the subtropical and subpolar ocean gyres.
They have found an opposite polarity located off the southern United States. The time
series of EOF 2 exhibits quasi-biennial fluctuation and also quasi-decadal variations. We
have also found similar results in our EOF and also the time series is similar after 1948 to
onward.
69
Fig 4.4: EOF analysis of SST from 1900-1989. (Deser & Blackmon)
There is also a possibility that interannual and decadal fluctuations in SST are response to
ocean and atmospheric variations outside of North Atlantic. Deser and Blanckmon,
(1993) have found a link with the sea ice transport in the Davis Strait-Labrador Sea
region. Figure 4.5 a) shows the time series winter sea ice concentration anomalies in the
Davis Strait-Labrador Sea. The circle denotes the winter anomaly, plotted in the year in
which January occurs. The solid curve shows the data smoothed with a three point
binomial filter. Decadal variability is evident in the sea ice record, with peaks occurring
in the winters of 1957/58, 1971/72, and 1983/84. Figure 4.5 b) shows the sea ice record
superimposed upon the time series of the second EOF of winter SST. ( Deser &
Blackmon, 1993)
It may be seen that the maxima in sea ice concentration precede the minima in SST by 1-
2 years. There is a strong lag in correlation between the two time series. This indicate that
on the decadal time scale, winters of heavy sea ice in the Labrador Sea precede winters of
colder than normal SSTs east ofNewfoundland. It is probable that the sea ice anomalies
70
in the Labrador Sea are advected southwestward, resulting in colder than normal SSTs
east of Newfoundland the following year. Thus the quasi-decadal cycle in SSTs east of
Newfoundland may result in part low-frequency Arctic sea ice variations.
40
20
.... 2 0
,, :·
f\1 :·t\r.l.
J \.'
-2~9~~~~~~~~~~~~~~80~~ ~
b
{b)
Fig 4.5 a) Time series of Winter Sea ice area anomalies in the Davis Strait and Labrador
Sea region. b) Standardized sea ice anomalies from super imposed upon the standardized
time series of EOF2 of North Atlantic winter SST (after Deser and Blackmon, 1993)
The dominant EOF mode of wind (Fig 4.6 a) explains 46% of the variance and the
second EOF explains 24% of the variance. The dominant EOF defines a pattern which
has a dipole structure with opposite variability in the sub-polar and sub-tropical regions.
This EOF, which dominates the zonal wind variability, explains in particular the
anticorrelation between the winds anomalies southern and northern of 45~. The pattern
related to the EOFl constitutes oscillations with opposite phase in these two regions. It's
time variability follows closely the NAO index.
The second EOF in the wind defines a triple pattern. The variability related to this EOF in
the wind in the mid-latitude area between 30~ and 55~ is opposite to that in the
subpolar and tropical North Atlantic.
71
" " * H
• -~
--.... _ ·--* N . • --.... -
.
010 __ , ___ ,
·- - ·- ·-"'.,__ .. __ COOl'. . .
·- • ·-•r------------~~,._,-,=~ .. --cOl''
• ... -
- -·
-
., .. ~---, ..... -.---,HM .. .---,,c,,.H;---;.,.-..---,, ..... ---. ..... .---.1 ...
Fig 4.6: Dominant EOFs of Wind and their lime series (red cun-e is I he jillered time
J·eris) of North Atlamic.
n
The first EOF of precipitation (Fig 4.7 a) of North Atlantic which explainsl6% of the
variance has highest precipitation rate in the northern coast of Newfoundland and US
coast and lowest in the southeastern part of North Atlantic. The second EOF, which
explains 14% of variance has highest precipitation rate in Gulf Stream and lowest in
central Atlantic. The third EOF of which explains 7% of the variance has highest
precipitation all over the Atlantic.
The first three EOFs in total precipitation explain only about 40% of the total variance.
The rest of the variance in the precipitation anomalies is due to the higher EOFs. In this
way, we need relatively large number of EOFs in order to spawn the precipitation
variability. The first EOF coefficient has a time variability, which follows closely the
variability of the SLP EOF1 coefficient. This is the pattern closely related to the NAO.
The dominant EOF mode of total heat flux explains 21% of the variance and the second
EOF explains 15% of the variance. The third structure explains 8% of the variance. Fig
4.8 shows the dominant EOFs for the surface net turbulent flux. The first EOF (Fig 4.8 a)
defines a pattern of alternating anomalies in the areas of subtropical and subpolar ocean
gyres. The low-pass filtered time variability of this EOF follows closely the time
variability of the first SLP EOF defining the NAO pattern. The second EOF defines a
pattern in the heat flux with alternating anomalies in the region southeastern of the
Newfoundland and the rest of the North Atlantic.
73
.. , .. ---- -· (.ol) T - ~lloeoft llQIO 1(0040.1~7:3)
' • •
' f ,~
• ·- . . . ..:.. ·- - "" ·- ·- -· ..... ___ ,_ -· • •
•
• ~ .,_ - ·- - ·- --- - -... ,,. ~,__ ___ .... ,I ..
.. •
••
•• . 1.:.. - ·- - ·- - - •
Fig 4. 7: Domlmmt EOF of Preclpllallon Rate and their time J·erles(retl curt't iJ the
fi/t~n-~1 wne series) of Atltmnc.
74
.,__, ____ . --• --- - - - - - - -
.___.., ____ , 1 ---
• v --- - - - - - - -t•) ~ ..... lbl: EOF 3 (-O.Dit0012) - ___ .. ____ .
--·-• --- - - - - - - -
FIR 4.8: /)Qminant EOFs of Total lleot J.1u.Yand their time serle\' (reel curve i.v the
filteml time serle<) of Atlantic.
7S
These results correspond to the findings of Cayan (1992). The total heat flux differences
associated with positive versus negative extremes of the first two North Atlantic SLP
EOFs are shown in Fig 4.9 and corresponding MST' I M differences are shown in Fig
4.1 0, which is the quantity of thermal energy transferred across a unit area of Sea Surface
per unit time. SST anomalies fall in regions of positive sea-to-air flux and rise in regions
of negative flux. In concurrence with the NAO circulation pattern, the North Atlantic is
partitioned into alternating anomalous cells: 1) positive MST' I M differences in the
northeastern North Atlantic between Great Britain and Scandinavia; 2) negative
!).SST' I M differences in the central North Atlantic southeast of Greenland between 50~
and 60~; 3) a broad region of positive MST' I M differences extends from the Gulf of
Mexico and the eastern seaboard east to the central Atlantic at about 350N,300W; 4)
negative MST' I M differences occupy the eastern tropical North Atlantic between about
500W and North Africa. The major regions of impact of MST' I M are closely
harmonized to those of total heat flux, supporting the view that they are linked.
For the North Atlantic SLP EOF2 (EA), the major MST' I M features are 1) negative
differences in the central North Atlantic east of Newfoundland; 2) positive differences in
the Gulf of Mexico and the region just east of Florida and 3) positive differences in the
far northeastern Atlantic and the Mediterranean Sea.
There is a difference in the pattern defined by the EOF2 of heat flux (Fig 4.8c) and the
corresponding pattern of Cayan (1992). The pattern on (Fig 4.8c) is shifted south
westward and is more extended towards the US coast in the Southern part. This
difference may be attributed to the different method used by Cay an ( 1992), which
considers the heat flux distribution during years of extremely high or extremely low
76
NAO. In this way the patterns of Cayan (1992) represent the structure of most anomalous
(in terms ofNAO index) years.
Fig 4.9: Difference of average total heat flux associated with positive vs negative extreme
winter month ofNorthAtlantic SLP EOFJ and 2 of (Cayan 1992)
77
Fig 4.10: Difference of average thlST' I M associated with positive vs negative extreme
ofEOFJ and 2 ofNorthAtlantic SLP. (Cayan 1992)
78
The dominant EOFs in the near surface atmospheric characteristics and SST define
patterns show close connection to the global atmospheric circulation. The EOF analysis,
for the period 1948-2005 in correspondence with previous studies, indicates that the
NAO has an impact on the North Atlantic Ocean and atmospheric variability. The EA is
another important pattern for the North Atlantic. At the same time it is known that
internal ocean variability may also influence the atmosphere and SST. The next chapter
studies the connection between the variability in the North Atlantic Ocean and Labrador
Sea, and the atmospheric circulation.
79
Chapter 5
Canonical Correlation Analysis
The anomalies of heat and water exchange between the atmosphere and the Labrador Sea
depend on the variability of wind speed, the ocean surface saturation humidity, air
humidity and the sea-air temperature difference. These parameters define the intensity of
evaporation and the net turbulent heat flux at the ocean-atmosphere interface. Their local
values depend on the atmospheric circulation, which defines the intensity of the surface
wind speed and the direction of air masses transport. During the winter, the dominant
transport is directed from the continent towards the ocean and brings cold and dry air to
the ocean surface. The heat and water exchange depends on the intensity of this flow,
which is part of the atmospheric circulation
This section presents results from a study of the relation between the atmospheric
circulation, and near surface atmospheric variability and SST. The time variability of SLP
dominant EOFs discussed in the previous section show some qualitative relation to the
EOFs in other atmospheric characteristics (zonal wind, heat flux, atmospheric
temperature) and SST. This relation was shortly discussed in previous section. For
instance, negative SLP anomalies are related to negative zonal wind anomalies to their
north and positive zonal wind anomalies to their south. The analysis of this relation in
chapter 4 was just qualitative. Here the method of canonical correlation analysis is used
to study the relation between the near surface atmospheric characteristics and SST with
the atmospheric circulation variability. Section 5.1 presents results for the North Atlantic.
80
The relations of the near surface atmospheric characteristics and SST in the Labrador Sea
to the atmospheric circulation is discussed in section 5.2.
5.1 Correlation Patterns of Variability in the North Atlantic
Canonical Correlation Analysis (CCA) is used to analyze the relationship between
monthly mean sea level pressure (SLP) or sea surface pressure (SSP) and atmospheric
temperature, wind, heat flux, SST, precipitation in the North Atlantic. The data are time
series of monthly means of SLP and atmospheric temperature, wind, heat flux, SST,
precipitation grided over the North Atlantic north of about 20~, for DJF of 1948 to 2005
respectively. Anomalies are obtained at each grid point by subtracting the long term
monthly mean from the original values. The CCA yields two pairs of patterns that
describe the coherent variations in the two parameters.
The first pattern (Fig 5.1 b) of SLP and SST corresponds to a canonical correlation of
84% that means the areas of strong coupling of SLP with SST indicated by 84%. The
coefficient of the second pattern (Fig 5.1d) is 68%. The coefficient of the third pattern
(Fig 5.1 f) is 58%. The first pattern (Figs 5.1 a, b) of SLP exhibits the pattern similar to
NAO pattern which drives the variability in the SST close to the 2nd EOF and second
mode of canonical correlation pattern of SLP describes EA pattern which drives the
variability in the SST close to the first SST EOF. These results correspond to the studies
of Cayan., (1992) and Deser et al., (2000). These studies in particular relate the second
EOF (Fig 4.5), or triple pattern in SST anomaly field with the NAO. The first EOF in
SST, defines a pattern that is close to the one derived by Cayan (1992) and driven by
Eastern Atlantic pattern in the atmospheric circulation.
81
The third correlation pattern of SLP and SST (Figs 5.1 e, f) has a SLP structure close to
the NAO but with a maximum in the SLP field shifted westward. This canonical
correlation pattern in the SST has a dipole structure with opposite variability in the
subtropical and subpolar gyres.
The first pair of patterns SLP and atmospheric temperature corresponds to a canonical
correlation of 90% of the SLP and air temperature. The correlation coefficient of the
second pairs of patterns is 85%. First canonical pattern (Figs 5.2 a) of SLP is North
Atlantic Oscillation (NAO) and drives the variability in the air temperature close to the
first EOF. The second mode (Fig 5.2 c) of canonical correlation is a combination between
NAO and Eastern Atlantic (EA) pattern that drives an air temperature anomaly close to
the second EOF. These two patterns show how the variability in atmospheric circulation
influences the near surface atmospheric temperatures as discussed in Chapter 1. High
NAO index is related to an intensified cyclonic vortex in the Icelandic low. The mean
transport in the Labrador Sea favors north-eastern cold winds. The winter storms are
passing across the Atlantic Canada and cause low temperature off Newfoundland. The air
temperature off the coast of US is warmer than usual.
The shift of the centers of low northward of Iceland and of the high northwestward of
Azores Islands in the NAO shifts the areas of high and low temperature in the second
canonical correlation patterns northward (Fig 5 .2d).
The first pattern of SLP and zonal wind (Fig 5.3 b) corresponds to a canonical correlation
of99.97%. The coefficient ofthe second pattern (Fig 5.3 d) is 99.90%. The coefficient of
the third pattern (Fig 5.3 f) is 99.61%.
82
The first two dominant SLP-zonal wind patterns (Figs 5.3 a, c) are related with the NAO.
In the first pattern the high in Eastern Atlantic is shifted northward of the Azores Island.
In this case the zonal wind is intensified in the eastern part ofNorth Atlantic off the coast
of England. The zonal transport is weaker than normal in the subtropical North Atlantic
Ocean.
The second SLP pattern (Fig 5.3 c) has a structure of the NAO with the low shifted south
of Iceland. Correspondingly the area of intensified zonal winds moves southward with
respect to the first pattern. The strongest zonal transport related to this pattern is between
Newfoundland and Western Europe. The third canonical correlation pattern (Fig 5.3 e)
has a structure of the Eastern Atlantic pattern in the SLP. It is known (Barnston and
Livezey, 1986) that the low related toEA pattern can move zonaly in the North Atlantic.
In canonical correlation pattern three, this center is shifted westward with respect to EOF
2 of SLP. During the positive phase of this pattern, the zonal wind intensify in the
tropical North Atlantic and off the west coast ofEurope.
The first patterns of SLP and heat flux (Fig 5.4 b) corresponds to a canonical correlation
of99.8%. The coefficient ofthe second pattern (Fig 5.4 d) is 99.45%. The coefficient of
the third pattern (Fig 5.4 f) is 99.09%. The fourth pattern of SLP and heat flux
corresponds to a canonical correlation of 98%. First canonical pattern of SLP is NAO
which drives the variability in the turbulent heat flux close to the first EOF of the heat
flux. The second and third SLP modes (Fig 5.4 c, e) of canonical correlation are similar
to the Eastern Atlantic (EA) pattern which drives the turbulent heat flux anomalies close
to the second EO F.
83
The first patterns of SLP (Fig 5.5 b) and precipitation corresponds to a canonical
correlation of 9 5% of the variance of SLP and wind. The coefficient of the second pattern
(Fig 5.5 d) is 93% of the variance of SLP and precipitation in the North Atlantic. The
coefficient of the third pattern (Fig 5.5 f) is 92% of the variance of SLP and heat flux in
the North Atlantic. The fourth pattern of SLP and precipitation corresponds to a canonical
correlation of 87% of the variance of SLP and precipitation. First canonical pattern of
SLP (Fig 5.5 a) is North Atlantic Oscillation (NAO) which drives the variability in the
precipitation close to the first EOF. The second mode (Fig 5.5 c) of canonical correlation
is similar to the Eastern Atlantic (EA) pattern, which drives the precipitation anomaly
close to the second EOF.
84
·- - - -
Fig 5.1: Upper two panels .show tltejir,fl Canonical L'Orre/alion J)llllernfor (t1) SLP or
SSP (b) SST 11/tld/c two panels show th<t second CC pattern/or (c) SLP (d) SST. f.ou·er
tuo panels showth<t third CC poiternfor (e} SLP (d) SST.
cc.- t: ... _OcoonSLP-
.. ... ·U 0 0.1
Fig 5.2. UpfJeT two /l('nel< show I he firs/ CC /l('llemfor (a) SLP (b) air lemperlllllre.
Lower two panels show the second ('("pattern/or (c) SI~P (d) llir temperature
86
Fig 5.3: Upper lwo panels show the firs/ Canonical correlalion p<lllem}Or (a) SLP (b)
zona/wind J\tliddle lwo fXInels show the second CC pallernfor (c) SLP (d) zonal wind
Lower lwo JXmels show the third CC fXlllem}Or (e) SLP (d) zonal wiml.
87
co-, ...... ,,,. ..... ._.,.._..
•
Fig S.4. Upper r.·o pun</$ shou IM fiT!/ Cononico/ correlallon pull<rn/Ot" (a) S/.P (b)
ltet~tfl•u:. Middle two panels show the ;econd CC pullernfor (c) SLP (d) heat flux. Lower
r.•o panels show the third CC pallernfor (e) SLP (d) hcOij/liX.
II
oo_., _ _ _
oo-·-- ..-.
Fig S.S ~"pfwr lln> pan< Is show tlw fmt Canonirol corr./Diion pallmr/or (a) SLP {b)
fNrclpilution rot~. MilitO~ n.o ponrls show t- s~rond C( pilll~rnfor (c) SLP (d)
precipitation rate. Lower t-..·o pcmt!l.J show the third CC palftrnfor (e) SLP (d)
J'redplfation rate.
19
5.2 Correlation Patterns of Variability in the Labrador Sea
In this section CCA is used to study the relationship between monthly mean sea level
pressure (SLP) North Atlantic and atmospheric temperature, wind, heat flux, SST,
precipitation in the Labrador Sea. The data are time series of monthly means of SLP and
atmospheric temperature, wind, heat flux, SST, precipitation on a grid over the Labrador
Sea north of about 50~, for DJF of 1948 to 2005 respectively. Anomalies are obtained at
each grid point by subtracting the long-term monthly mean from the original values. The
CCA yields two pairs of patterns that describe the coherent variations of the atmospheric
temperature, wind, heat flux, SST and SLP fields and three patterns that describe
precipitation with the SLP fields. These patterns are dominant in describing SLP with
other parameters variance in the Labrador Sea.
The first pattern of SLP and SST (Fig 5.6 b) corresponds to a canonical correlation of
85%. The coefficient of the second pattern is (Fig 5.6 d) 72%. First canonical pattern of
SLP is North Atlantic Oscillation (NAO) which causes cold temperature in the Labrador
Sea with strongest impact in the northern part of the Labrador Sea and some part in the
southern area. The second mode of canonical correlation is similar to the Eastern Atlantic
(EA) pattern which has strongest impact on the northern part of the Labrador Sea.
The canonical correlation patterns are obtained through linear correlation analysis of SLP
and SST fields. The dominant patterns of CCA are well correlated in time and therefore
one can use them to predict the values of SST if SLP is known. Prediction of SST is
based on using canonical variates for SLP to project variability of SLP into
corresponding canonical correlation patterns of SST. The reconstruction of SST by using
SLP canonical correlation variates is given by Zorita et al (1992):
90
R " (t ) = " .f3 SLP (t ) fSST S5l n L..,. P1 1 n J 1 5.1
This approach is used in statistical weather forecasting methods. Some problems may
arise when the CCA is used to reconstruct one parameter (SST) from known values of
another (SLP). CCA patterns (Zwiers and von Storm, 1999) can not necessarily explain
all the variance of the SST. Following the terminology of Chapter 2, let us consider the
two parameters SST = SST d+SST', SLP = SLP d + SLP', each consisting of two
components - deterministic and stochastic. The canonical correlation analysis is based on
the correlation between SLP d and SST d represents variabilities in the two parameters
which are not correlated. The variance of SLP' and SST' define importance of factors
which are not considered by the CCA.
In the case of SST and SLP, the basic assumption of CCA is that atmospheric circulation
(SLP) variability influences the surface heat fluxes in the Labrador Sea, which are
important for the time changes of temperature in the surface ocean layer. Bjerknes., 1964
suggested this assumption is correct at interannual time scales, i.e. that interannual
changes in SST correlate well with the dominant modes of atmospheric circulation. This
is not the case, however, for the long term warming trends in the North Atlantic. In
chapter 3, it was shown that such trends are present in the North Atlantic in the 1950s and
late 1990s. Bjerknes (1964) suggests that the decadal trends of SST are related to the
variations of ocean circulation. The figure 5.7 shows water transport between Bermuda
and Bravo station. The green curve is NAO index filtered by 12 years and the black curve
is water transport and the red curve is SST anomaly. The NAO index (see Fig 5.7)
increase from 1980 to 1995. Water transport also increased during this period (Curry &
MacCartney, 2001 ). The SST trend in the 1990s is well correlated with the low pass
91
filtered NAO index and ocean transport. The major physical process responsible for this
long-term correlated trend in the three parameters is the ocean circulation. The transport
in the subtropical and subpolar gyre intensified after 1980 due to the NAO which is in
positive phase during the most of the time after 1980. The more intensive transport is
related to intensified heat transport in the surface ocean layer from the subtropics to the
subpolar region. Correspondingly the SST increases as a response to this transport. There
are two major properties of the SST, ocean transport and NAO variabilities during the
periods of positive SST trends, which are seen on Fig 5.7 and have impact on the quality
ofthe CCA and reconstruction of SST by using SLP:
1. The ocean transport responds with some delay to NAO. Moreover when the
meridional heat transport is intensified, during positive phase of NAO the time
required by a positive SST anomaly to propagate from the subtropical area to the
Labrador Sea is evaluated by Sutton and Allen (1997) as about 7-8 years.
n. The low-passed filtered variability in the NAO has variance much lower than the
variance ofNAO at interannual time scale see (blue and red bars on Fig 5.7)
The CCA patterns here do not consider an eventual time lag in the correlations between
the intensity of atmospheric circulation and SST. However if such time lag would be
considered, CCA will still not capture the correlation in the long term low pass filtered
trends between SST and SLP, because the filtered SLP trend has much lower variance
than the interannual variability in the nonfiltered SLP data.. Additionally the sea ice
variability in the Labrador Sea influences SST, introducing additional changes in SST,
which are not directly connected to the SLP variability. Fig 4.7 b shows the ice inflow to
92
the Labrador Sea through the Davis Strait has a strong impact on the dominant mode of
variability with a time lag of 1-2 years.
The CCA SLP-SST does not consider factors like ocean circulation, sea ice transport
which are important for the SST variability. Therefore one can expect that the
reconstruction of SST by using CCA patterns will not approximate perfectly the real SST
field. In order to assess the approximation skills of CCA reconstruction the following
parameter is used (Cayan, 1992):
(5.2)
which estimates the variance of SST explained by CCA patterns.
In Figure 5.6 (e) the blue line is the SST anomaly in the Labrador Sea and green line is
SST anomaly reconstructed by CCA. In this figure it is clearly shown that the SST is
weakly correlated with SLP and only in the cold period from 1970-1980s.
The variance explained by the CCA (Fig 5.6 f) is around 20% in the southern part of the
Labrador Sea. In the rest ofthe basin, the approximation skill is very negative. This result
shows that SST in the Labrador Sea is only weakly correlated with SLP patterns and the
variability of the atmospheric circulation. Strong impact on the SST in the Labrador Sea
have ocean heat and sea ice transport.
93
•• r--- _.,.. ___ .-T __ ._
•• •• •• .. •
••
·- ·- ·- - -· Fig 5.6: Upper two panels show the first (.(monica/ correlation pallemfor (a) SI.P (b)
SST. Middle two panels .vlrow tire second CC pallernfor (c) SLP (d) SST. e) Lower two
panels slrow SST anomaly (blue curve) and reconstructed SSt anomaly (green curve)
d) Variance of SST(%) explained by CC reconsrruclion
- bl!plrti'Qa 0-md! - ..... N() .... d!J'I
·-""' --Fig 5.7: Water transport index, intt~rated NAO inde:c (Curry and Mac:Cttrftrey. 2001)
and SST anomalies (Sl111on, Allen 1997)
The first pairs of pottems of SLP and atmoophcric tempen~tun: correspond$ to a canonical
correlation of 98o/o. ·n1e cocllicient or the second pairs of patterns is 96~o. First canonical
pottem ofSLP (l'i& S.8 a) which has strongest impact in the northern part of the Labrador
Sea is lhe No<th Atlantic Oscillation (~AO). The second mode of canomcal (fig S.8 c)
which has stronge:>t impact also in the nonhem part of the l.nbrador Sen i!; similar to the
Enstem Atlantic (EA) patlem. 11tc SLP and air tempcrnture in the Labmdor Sea are well
correlated. The variance explained by lhe CCA (Fi& S.8 f) is around 60''1.-80%
cvc;ry~·here except cc:nlral po.n of the Labrador Sea ~here the explained \ariance is
around 40'/o.
9S
The first patterns of SLP and wind corresponds to a canonical correlation of 99%. The
coefficient of the second pattern is 97%. First canonical pattern of SLP (Fig 5.9 a) which
has strongest impact in the southern part of the Labrador Sea and near Greenland is North
Atlantic Oscillation (NAO). The second mode of canonical correlation (Fig 5.9 c) which
has strongest impact near the Greenland coast is similar to the Eastern Atlantic (EA)
pattern. The SLP patterns and zonal wind in the Labrador Sea are well correlated. The
variance explained by the CCA (Fig 5.9 f) is around 80%-100% everywhere except in the
Davis Strait and some costal parts ofNewfoundland where no correlation is found.
The first patterns of SLP and turbulent heat flux corresponds to a canonical correlation of
97%. The coefficient of the second pattern is 95%. First canonical pattern of SLP (Fig
5.10 a) which has strongest impact in the southern part of the Labrador Sea and near
Greenland is North Atlantic Oscillation (NAO). The second mode of canonical
correlation (Fig 5.10 c) which has strongest impact also in the southern part of the
Labrador Sea is similar to the Eastern Atlantic (EA) pattern. The SLP and turbulent heat
flux pattern patterns in the Labrador Sea are well correlated. The variance explained by
the CCA (Fig 5.10 f) is around 60%-80% everywhere except Davis Strait, Hudson strait
and coastal part ofNewfoundland.
The first patterns of SLP and precipitation corresponds to a canonical correlation of 98%.
The coefficient of the second pattern is 97%. The coefficient of the third pattern is 96%
of the variance of SLP and heat flux in the Labrador Sea. First canonical pattern of SLP
(Fig 5.11 a) which has strongest impact in the northern part of the Labrador Sea and near
96
Greenland is North Atlantic Oscillation (NAO). The second mode (Fig 5.11 c) of
canonical correlation which has strongest impact also in the coastal part of Greenland is
similar to the Eastern Atlantic (EA) pattern. The SLP and air temperature in the Labrador
Sea are well correlated. The variance explained by the CCA (Fig 5.11 f) is around 60%-
80% everywhere except northern part of Labrador Sea and coastal part ofNewfoundland.
97
• r _;·-:!'!:' =-~·-~·:!~~-=-~::·~-- __ .. _ • •
• -· • •
Fig 5.8: Upper MO pone/s show the first C<monical correlation p(l{ternfor (a) SLP (b)
air temperature. Middle two panels show the second CC potternfor (c) Sl.P (d) air
temperature. e) Lower two panels show air temper(llure cmomaly (blue curve) and
reconstructed air temperature anomaly (green curve) d) Variance of air temperature rAJ
explained by CC reconstruction
98
r __ ,.:,..::•;:•:::-:::-;::::"-=-:=::-=::";:'='::-::;..:•::"=-~--, '
••
•• •
•• ·• ... •
I
' I'
fig 5. 9: Upper IWO ponels show the first Canonical correlation polternfor (a) SLP (b)
zonal wind. Middle two ponels show the second CC pattern for (c) SLP (d) zonal wind.
e) Lower two panels show zonal wind anomaly (blue curve) and reconstructed zonal wind
anomaly (green curve) d) Variance of zonal wind(%) explained by CC recon.slruction
~
y
M
• • • ... ... . ,...
~-----.......,-·-
....
Fig 5./0: Upper two ponels show the first Canonical correlation pollemfor (a) SU' (b)
heatji1L<. Middle rwo poneL< show the second CC pallemfor (c) SLP (d) heatj/1<r.
e)Lou:er two panels show heat flux anomaly (blue curve) and reconslructed heal flux
anomaly (green curve) d) Variance ofhealjlux r/oJ explained by CC reconslrucliOn
100
cc mode 1: the Lab s- Prate pattem CC mode 1: the AUantJc Ocean SSP pattem
CC mode 2: the Lab Sea Prate pattem CC mode 2: the AUantic Ocean SSP pattem
x10-e
1 x10-e Variance (%) explaind by the CCpatems
0.8
0.8
0.4
0.2
0
.0.2
.0.4
.0.8
.0.8
-1940 19!50 11180 1970 1980 1990 2000 2010
Fig 5.11: Upper two panels show the first Canonical correlation pattern for (a) SLP (b)
prate_ Middle two panels show the second CC pattern for (c) SLP (d) prate. e)Lower two
panels show precipitation anomaly (blue curve) and reconstructed precipitation anomaly
(green curve) d) Variance of precipitation (1/o) explained by CC reconstruction
101
Chapter 6
6. Conclusions and Future Work
EOF and CCA are techniques commonly used in analysis of geophysical data. In this
study the application of these two statistical methods are used to study interannual and
interdecadal variability of Labrador Sea. The purpose of this study is to find the dominant
modes of variability in the North Atlantic SST and near surface atmospheric parameters
and to asses the effect of the atmospheric circulation on the SST and the atmospheric
parameters in the Labrador Sea and the North Atlantic.
• The SST and near surface atmospheric parameters are studied for the period of
time 1950- 2005.
• The data used m this study is Comprehensive Ocean-Atmospheric Dataset
(COADS). The SST data are used from this COADS dataset. And the other
parameters are taken from NCEP reanalysis.
• The dominant modes in the North Atlantic interannual variability are found by
using EOF analysis
• The dominant modes of interannual variability in the Labrador Sea and their
relation to atmospheric circulation are studied by using CCA.
• The approximation skills of canonical correlation patterns of SST, zonal wind,
turbulent heat flux, air temperature and precipitations are computed. The CCA
SLP-SST does not consider factors like ocean circulation, sea ice transport that
102
are important for the SST variability. Therefore it can expect that the
reconstruction of SST by using CCA patterns will not approximate perfectly the
real SST field. In order to assess the approximation skills of CCA reconstruction
the following parameter is used (Cayan, 1992):
Ji = [1- Lk (RSST (t n)- SST(t n )) 2 / Lk (SST(t n )) 2
] * 100%
which estimates the variance of SST explained by CCA patterns.
• The winter SST anomaly in the Labrador Sea especially during warm periods
does not show significant correlation with the patterns of atmospheric circulation.
• This result supports Bjerknes (1964) conclusion that decadal variability of SST
during warm periods depend mostly on non-local processes of horizontal transport
and less on the local heat exchange with the atmosphere. Inter-decadal warming
of SST in the North Atlantic is linked to the basin scale interaction. The Gulf
Stream and the North Atlantic current respond to the intensifying circulation in
the anti-cyclonic gyre.
• The near surface atmospheric parameters show good correlations with SLP.
From the results presented in Chapter 5 it follows that the future analysis of the climate
variability in the Labrador Sea should be based on additional data about variability of the
Atlantic Ocean. The future work would include the following steps:
• Consideration of the ocean circulation and its long term effect on SST and
atmospheric parameters.
• Using of realistic ocean model simulations and observations.
• Application of the EOF and CCA for both atmospheric characteristics and ocean
surface and sub-surface parameters.
103
Abbreviations
• NAO: North Atlantic Oscillation. The NAO is the dominant mode of winter
climate variability in the North Atlantic region ranging from central North
America to Europe and much into Northern Asia. The NAO is a large scale
seesaw in atmospheric mass between the subtropical high and the polar low. The
corresponding index varies from year to year, but also exhibits a tendency to
remain in one phase for intervals lasting several years. The Positive NAO index
phase shows a stronger than usual subtropical high pressure center and a deeper
than normal Icelandic low. The increased pressure difference results in more and
stronger winter storms crossing the Atlantic Ocean on a more northerly track. The
negative NAO index phase shows a weak subtropical high and a weak Icelandic
low. The reduced pressure gradient results in fewer and weaker winter storms
crossing on a more west-east pathway.
• NCEP: National Center of Environmental Prediction.
• COADS: Comprehensive Ocean-Atmospheric Dataset.
104
• EOF: Empirical Orthogonal Function. The method of empirical orthogonal
function (EOF) analysis is a decomposition of a signal or data set in terms of
orthogonal basis functions which are determined from the data.
• CCA: Canonical Correlation Analysis. A typical use for canonical correlation in
the general context is to take a two sets of variables and see what is common
amongst the two tests.
• SST: Sea Surface Temperature .
• SSP: Sea Surface Pressure .
• LSW: Labrador Sea Water .
105
Bibliography
Alexander A. M and Clara Deser (1995). The Reemergence of SST Anomalies in the
North Pacific Ocean, J Clim, 2419-2433.
Bamston, A. G., and R. E. Livezey, (1987). Classification, seasonality and persistence of
low frequency atmospheric circulation patterns, Mon. Wea. Rev., 115, 1083-1126.
Battisti D. S, Bhatt U. S, and Alexander M. A (1995). A Modeling Study of the
Interannual Variability in the Wintertime North Atlantic Ocean, J Clim, 3067-
3083.
Bjerknes, J ., (1964) Atlantic air-sea interactions, Advances in Geophysics, 10, 1-82.
Bretherton, C. S., Smith, C., and Wallancw, J.M. (1992). An intercomparison ofmethods
for finding coupled patterns in climate data. J Clim, 5,541-560.
Cayan, D. R., (1992a). Latent and sensible heat flux anomalies over the northern oceans:
Driving the sea surface temperature, J Phys. Oceano gr., 22, 859-881.
Cayan, D. R., (1992b) Latent and sensible heat flux anomalies over the northern oceans:
The connection to monthly atmospheric circulation, J Climate, 5, 354-369.
Curry, R. G., and M.S. McCartney, (2001). Ocean gyre circulation changes associated
with the North Atlantic Oscillation, J Phys. Oceanog., 31, 3374-3400.
Czaja A. and J. Marshall (2001). Observations of Atmosphere- Ocean coupling in the
North Atlantic, Quart. J of the Roy. Met. Soc., 127, 1893-1916.
Deser, C.,(2000). On the teleconnectivity of the "Arctic Oscillation." Geophys. Res. Lett.,
27, 779-782.
106
Deser, C., and M. L. Blackmon, (1993). Surface climate variations over the North
Atlantic Ocean during winter: 1900-1989. J Clim, 6, 1743-1753.
Deser C., Walsh JE, Timlin MS .,(2000). Arctic Sea Ice Variability in the Context of
Recent Atmospheric Circulation Trends, J Clim, 617-633.
Dickson, R. R., J. Lazier, J. Meincke, P. Rhines, and J. Swift, (1996). Long-term-co
ordinated changes in the convective activity of the North Atlantic, Prog.
Oceanogr., 38,241-295.
Dickson R. Rand Namias J. (1976). North American Influences on the Circulation and
Climate ofthe North Atlantic Sector, Monthly Weather Review, 1255-1265.
Folland, C. K., and co-authors (2001), Observed climate variability and change, m
Climate Change 2001, The Scientific Basis, J. T. Houghton, Y. Ding, D. J. Griggs,
M. Noguer, P. J. van der Linden and D. Xiaosu, Eds., pp. 99-181, Cambridge
Univ. Press.
Frankignoul, C; Hasselmann, K (1977). Application to sea-surface temperature anomalies
and thermocline variability. Stochastic climate models. II, 29, 289-305.
G.R. North, T.L. Bell, R.F. Cahalan, and F.J. Moeng (1982). Sampling errors in the
estimation of empirical orthogonal functions, Mon. Wea. Rev., 110:699-706.
Golub, G. H., and Loan, C. F. V (1996). Matrix Computations (Third Edition), 476p.
Hansen, J., R. Ruedy, M. Sato, and K. Lo (2002), Global warming continues, Science,
295.
Hasselmann (1976) Stochastic climate models. Part I: Theory, Tellus, 28, 473-485.
Hayden B. P. (1981) Climate change and extratropical storminess in the United States: J
american water resourse association. 36, 1387-1397.
107
Hurrell, J. W., (1995). Decadal trends in the North Atlantic Oscillation: regional
temperatures and precipitation. Science, 269, 676-679.
Hurrell, J. W., and H. van Loon (1997). Decadal variations in climate associated with the
North Atlantic Oscillation, Climatic Change, 36, 301-326.
Jones, P. D., T. J. Osborn, and K. R. Briffa (2001). The evolution of climate over the last
millennium, Science, 292, 662-667,.
Kushnir, Y., W. A. Robinson, I. Blade, N. M. J. Hall, S. Peng, and R. T. Sutton (2002).
Atmospheric GCM response to extratropical SST anomalies: Synthesis and
evaluation, J Climate, 15, 2233-2256.
Kushnir, Y., (1994). Interdecadal variations in the North Atlantic Sea surface temperature
and associated atmospheric conditions. J Climate, 7, 141-157.
Lazier, J. R.N., 1980. Oceanographic conditions at Ocean Weather Ship BRAVO, 1964-
1974. Atmos.-Ocean, 18, 227-238.
Lazier J.R.N (1995). The salinity decrease in the Labrador Sea over the past thirty years.
Climate on Decade-to-Century Time scales, National Academy of Sciences Press,
295-302.
Malberg,H, Bokens,G (1997) Winter and summer temperatures in Berlin since 1929 and
their relationship with the North Atlantic Oscillation. Meteorologische Zeitschrift,
Vol.6, No.5, pp.230-242.
Mann, M. E., R. S. Bradley, and M. K. Hughes (1999). Northern Hemisphere
temperatures during the past millennium: Inferences, uncertainties, and
limitations, Geophys. Res. Lett., 26, 759-762.
108
Mitchell (1976). An overvtew of climatic variability and its cause & mechanism.
Quaternary Res, 6, 481-493.
Milankovitch M. (1941 ). History of radiation on the earth and its use for the problem of
the ice ages. K. Serb. Akad. Beogr. Spec. Pub!. 132. 633 pp.
Portis, D. H., J. E. Walsh, M. El Hamly, and P. J. Lamb (2001). Seasonality of the North
Atlantic Oscillation, J Climate, 14, 2069-2078.
Rogers J. C. (1990). Patterns of low-frequency monthly sea-level pressure variability
(1899-1986) and associated wave cyclone frequenciesJ Clim, 10, 1635-1647.
Rogers J. C. (1997). North Atlantic Storm track variability and its association to the
North Atlantic Oscillation and climate variability of northern Europe, J. Climate,
10, 1635-1647.
Sutton, R. T. Allen, M. R. (1997). Decadal predictability of North Atlantic sea surface
temperature and climate. Nature, 388, 563-567.
Talley L.D (1984). Meridional heat transport in the Pacific Ocean, J Phys. Oceangr. 14
231-241.
Thompson, D. W. 1., and J. M. Wallace (2000). Annular modes in the extratropical
circulation, Part I: Month-to-month variability, J Climate, 13, 1000-1016.
Van Loon, H., and J.C. Rogers (1978). The seesaw in winter temperatures between
Greenland and Northern Europe. Part I: General description. Mon. Wea. Rev.,
106,296-310.
Visbeck, M., H. Cullen, G. Krahmann, and N. Naik, (1998). An ocean model's response
to North Atlantic Oscillation-like wind forcing. Geophys. Res. Lett., 25, 4521-
4524.
109
Wallace, J. M., Smith, C., and Bretherton, C. S. (1992). Singular value decomposition of
winter time sea surface temperature and 500-mb height anomalies. J Clim., 5,
561-576.
Wallace and Gutzler D. S. (1981), Teleconnections in the geopotential height field during
the Northern Hemisphere winter, Man, Wea, Rev., 109, 784-812.
Woodruff, S.D., R. J. Slutz, R.L. Jenne, and P.M. Steuer, (1987). A compherensive
ocean-atmosphere dataset. Bull. Amer. Metor. Soc, 68, 1239-1250.
Zorita. E., Kharinv., and H. von Storch, (1992). The atmospheric circulation and the SST
in the North Atlantic area in winter: Their interaction and relevance for Iberian
precipitation. J Clim., 5, 1097-1108.
110